CN109902957B

CN109902957B - Data processing method and device

Info

Publication number: CN109902957B
Application number: CN201910150618.9A
Authority: CN
Inventors: 闫颖异; 雷欣翔; 陈平
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2019-02-28
Filing date: 2019-02-28
Publication date: 2022-12-09
Anticipated expiration: 2039-02-28
Also published as: CN109902957A

Abstract

The invention discloses a data processing method and a device, wherein the method comprises the following steps: obtaining target text information from the obtained target service text data, and carrying out abnormal detection on the target text information according to a target quality inspection point in a target quality inspection rule; if the target text information has target quality inspection points meeting the target quality inspection rule, setting a first identification corresponding to the target quality inspection points in the target text; acquiring associated text information associated with the target text information from the target service text data, and performing anomaly detection on the associated text information according to associated quality detection points in the target quality detection rule; if the associated text information has the associated quality inspection points meeting the target quality inspection rule, setting a second identifier corresponding to the associated quality inspection points in the associated text information, and determining target service text data carrying the first identifier and the second identifier as a target abnormal text. By adopting the embodiment of the invention, the workload of quality testing personnel can be reduced, and the quality testing efficiency is improved.

Description

Data processing method and device

Technical Field

The present invention relates to the field of computer technologies, and in particular, to a data processing method and apparatus.

Background

In the field of customer service quality inspection, service voice data and interactive text data are continuously increased, and higher requirements are put forward on the data processing speed in the customer service quality inspection process.

In the prior art, in the quality inspection process, a large amount of service work orders (namely text data obtained when customer service interacts with a user) can be sampled in a manual sampling inspection mode, so that each detail item with abnormality is found in the sampled service work orders in a manual error checking mode for scoring, and therefore when a large number of sampling samples exist, the workload of quality inspectors can be greatly increased.

In addition, the existing quality inspection process can also carry out preliminary screening on massive service work orders through a machine so as to select the service work orders with lower scores for the quality inspection personnel to carry out manual recheck. Because the grade of each service work order is obtained by the quality inspection through the machine, the quality inspector still needs to find out each error detail item from each service work order one by one in a manual error checking mode when performing secondary recheck, and the operation is complex, so that the quality inspection efficiency is low.

Disclosure of Invention

The embodiment of the invention provides a data processing method and device, which can reduce the workload of quality inspection personnel and improve the quality inspection efficiency.

One aspect of the present invention provides a data processing method, including:

acquiring target service text data, acquiring target text information from the target service text data, and performing anomaly detection on the target text information according to a target quality inspection point in a target quality inspection rule;

if the target text information has a target quality inspection point meeting a target quality inspection rule, setting a first identifier corresponding to the target quality inspection point in the target text;

acquiring associated text information associated with the target text information from the target service text data, and performing anomaly detection on the associated text information according to associated quality inspection points in the target quality inspection rule;

if the associated text information has associated quality inspection points meeting the target quality inspection rule, setting a second identifier corresponding to the associated quality inspection points in the associated text information, and determining target service text data carrying the first identifier and the second identifier as target abnormal text.

One aspect of the present invention provides a data processing apparatus, including:

the text information acquisition module is used for acquiring target service text data, acquiring target text information from the target service text data and carrying out exception detection on the target text information according to a target quality inspection point in a target quality inspection rule;

the first setting module is used for setting a first identifier corresponding to a target quality inspection point in the target text if the target text information has the target quality inspection point meeting a target quality inspection rule;

the associated information acquisition module is used for acquiring associated text information associated with the target text information from the target service text data and carrying out abnormal detection on the associated text information according to associated quality inspection points in the target quality inspection rule;

and the second setting module is used for setting a second identifier corresponding to the associated quality inspection point in the associated text information if the associated quality inspection point meeting the target quality inspection rule exists in the associated text information, and determining target service text data carrying the first identifier and the second identifier as a target abnormal text.

One aspect of the present invention provides a data processing apparatus, including: a processor and a memory;

the processor is connected to a memory, wherein the memory is used for storing program codes, and the processor is used for calling the program codes to execute the method according to an aspect of the embodiment of the invention.

An aspect of an embodiment of the present invention provides a computer storage medium storing a computer program, the computer program comprising program instructions that, when executed by a processor, perform a method as in an aspect of an embodiment of the present invention.

The method comprises the steps of obtaining target text information in target service text data, determining whether a target quality inspection point exists in the target text information according to a target quality inspection rule, if so, setting a first identification corresponding to the target quality inspection point in the target text information, further obtaining associated text information associated with the target text information, determining whether the associated text information has the associated quality inspection point according to the target quality inspection rule, if so, setting a second identification corresponding to the associated quality inspection point in the associated text information, and further determining the target service text data carrying the first identification and the second identification as a target abnormal text. Therefore, in the whole quality inspection process, the fine items with errors can be quickly found out from the target service text data through the obtained target quality inspection rule, the found fine items are marked to obtain the first identification and the second identification, and the target service text data with the marks can be determined as the target abnormal text. In other words, the target abnormal text carries the first identifier and the second identifier, so that the subsequent quality inspection personnel can quickly locate error details according to the marks, the workload of the quality inspection personnel is reduced, and the quality inspection efficiency is improved.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.

Fig. 1 is a schematic view of a data processing method according to an embodiment of the present invention;

fig. 2 is a schematic flowchart of a data processing method according to an embodiment of the present invention;

FIG. 3 is a flow chart of another data processing method provided by the embodiment of the invention;

FIG. 4 is a schematic diagram of a framework structure of a customer service quality inspection method according to an embodiment of the present invention;

FIG. 5 is a flow chart illustrating another data processing method according to an embodiment of the present invention;

fig. 6a and fig. 6b are schematic frame structures of another customer service quality inspection method according to an embodiment of the present invention;

FIG. 7 is a timing diagram illustrating a customer service quality inspection method according to an embodiment of the present invention;

FIG. 8 is a block diagram of a data processing apparatus according to an embodiment of the present invention;

fig. 9 is a schematic structural diagram of another data processing apparatus according to an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

Referring to fig. 1, fig. 1 is a schematic view of a data processing method according to an embodiment of the present invention. As shown in fig. 1, in the process of performing quality inspection on customer service, the server 100 may obtain session content between customer service and a user from each channel (e.g., online customer service, call center, etc.), that is, service voice data between customer service and the user, and the server 100 may input the obtained service voice data into a voiceprint recognition model, where the voiceprint recognition model is obtained by training according to a voiceprint database containing a voiceprint of customer service entered in advance and a common speech technique of customer service (e.g., "happy for your service"), and therefore, in the embodiment of the present invention, the trained voiceprint recognition model may be used to recognize a speaker. In other words, the voice content belonging to the customer service role can be recognized from the service voice data through the voiceprint recognition model, that is, the first text data corresponding to the customer service role can be identified from the service text data obtained by text conversion of the service voice data. Therefore, the remaining voice content in the service text data can be determined as the second text data of the user role, and the target text information can be determined according to the first text data and the second text data

The server 100 may perform voice recognition on the service voice data after the role distinction is performed, that is, the service voice data is converted into service text data, and the converted service text data may include the first text data and the second text data. It is understood that the speech recognition model is trained according to a large amount of speech data in the corpus database, and therefore, the speech recognition model can be understood as having a text conversion function. It should be understood that, in the embodiment of the present invention, the service voice data may be understood as voice data for characterizing a session between the customer service and the user in any one of a plurality of service channels.

After the service text data is obtained through voice recognition, the service text data can be input into an intelligent analysis engine in a quality inspection library, and the service text data is subjected to abnormal detection through a quality inspection model in the intelligent analysis engine, so that a target abnormal text can be determined according to an abnormal detection result. In other words, in the embodiment of the present invention, the service text data may be screened and marked by the quality inspection rule configured in the quality inspection model. In the process of the anomaly detection, the service text data without errors can be determined as normal texts, the service text data with errors which can be directly determined according to the quality inspection rule can be directly detected and determined as confirmed abnormal texts, and for the service text data which is detected as suspected to be erroneous, the quality inspection model can mark suspected error items of the service text data which is suspected to be erroneous, the service text data which is suspected to be erroneous is determined as target abnormal texts, the determined target abnormal texts can be added into the abnormal text data set, and the target abnormal texts are sent to the quality inspection terminal 200, so that the quality inspection personnel can manually review the received target abnormal texts, in other words, the quality inspection terminal 200 can perform secondary review according to the received service work orders (namely target abnormal texts) with the error item marking information, and then the service work orders with the abnormal items with errors can be determined from the service work orders which are suspected to be erroneous and sent to the anomaly server 100, and the server 100 can update the quality inspection rule according to the errors in the service work orders of the quality inspection personnel.

The quality control rule may be a rule that can be used to perform quality control on a large amount of service workflows (i.e., service text data obtained during a session between the client and the user), and the quality control rule may include a check formula obtained by combining check points with or without a check point. The check point may be referred to as a quality check point, and the check point may be data in a service link of any one of the massive service work orders, for example, the check point may be data in text data corresponding to the client role. The quality inspection model is a model that can perform anomaly detection on a large number of service work orders input into the quality inspection library, and can perform intelligent marking on an error item in a service work order with an error being confirmed and a suspected error item in a service work order with a suspected error, for example: deep neural networks, generating antagonistic networks, etc. The abnormal text data set may include all historical service text data determined as target abnormal text, that is, service work orders for storing suspected errors detected by quality inspection.

After performing a secondary recheck on the service work order (i.e., the target abnormal text) suspected of having the problem to obtain a quality inspection result, the quality inspection terminal 200 may send a quality inspection rule corresponding to the quality inspection result to the server 100. For convenience of understanding, the quality control rule configured in the server 100 may be used as the first quality control rule, and the quality control rule in the second review may be used as the second quality control rule. In other words, the quality inspection terminal 200 may feed back the second quality inspection rule to the server 100, and the server 100 may acquire an abnormal quality inspection point in the second quality inspection rule and update the first quality inspection rule.

Referring to fig. 2, fig. 2 is a schematic flowchart of a data processing method according to an embodiment of the present invention. As shown in fig. 2, the method may include:

step S201, obtaining target service text data, obtaining target text information from the target service text data, and carrying out abnormity detection on the target text information according to a target quality inspection point in a target quality inspection rule;

specifically, the server may select a service text data (i.e., interaction data between the customer service and the user) to be subjected to quality inspection from the quality inspection library as a target service text data, may obtain target text information according to different roles in the service text data, and analyze the target text information by using an intelligent analysis engine in the quality inspection library to determine whether the target text information has an abnormal condition, in other words, analyze the target text information by using a target quality inspection point in a target quality inspection rule configured in the intelligent analysis engine to determine whether the target text information hits the target quality inspection point. It can be understood that all service text data requiring quality inspection can be transmitted to the quality inspection library, and each time a quality inspection process is executed, one service text data can be selected from the quality inspection library as a target service text data, and then target text information is obtained from the target service text data, where the target text information may be not limited to specific content and quantity, and may be text content belonging to customer service in the target service text data (i.e. the words spoken by the customer service, which may be one or several words spoken by the customer service) as target text information, that is, text content belonging to the user in the target text data (i.e. the words spoken by the user, which may be one or several words spoken by the user) as target text information, or a section of text content containing continuous conversation between the customer service and the user; the target quality inspection rule refers to one quality inspection rule matched with a service scene to which the target service text data belongs in the multiple quality inspection rules, and the target quality inspection point can be a front inspection point in the target quality inspection rule, namely, the rest inspection points in the target quality inspection rule depend on the hit condition of the front inspection point. For example, in a service upgrade scenario, if a keyword "upgrade" occurs in the target service text data, the text content corresponding to the "upgrade" keyword (for example, 2 sentences before and after the "upgrade" keyword occurs) may be determined as the target text information, and the target quality inspection point in the target quality inspection rule may be "help you upgrade right" said by the customer service, or "please help me upgrade right" said by the user, and the like.

Optionally, in the intelligent analysis engine, a plurality of target service text data may be determined at the same time, and the plurality of target service text data may be analyzed in parallel, that is, each time a quality inspection process is performed, abnormality detection may be performed on the plurality of target service text data at the same time.

Step S202, if a target quality inspection point meeting a target quality inspection rule exists in the target text information, setting a first identification corresponding to the target quality inspection point in the target text;

specifically, the process of performing the abnormality detection on the target text information according to the target quality inspection point may include the following two cases: if the target text information has target quality detection points meeting the target quality detection rule, namely the target text information hits the target quality detection points, marking the target quality detection points in the target text information, wherein the marking information can be called as a first identification corresponding to the target quality detection points, so that an error item can be directly positioned according to the first identification during subsequent manual recheck; if the target text information does not have a target quality inspection point meeting the target quality inspection rule, that is, the target text information does not hit the target quality inspection point, the target service text data can be determined as a normal text without errors. The first identifier is used for marking the hit condition of the target quality inspection point, and may be represented by a numerical value or other characters. For example, in a service upgrade scenario, a target quality inspection point in a target quality inspection rule is an information keyword such as "help you upgrade right soon" spoken by a customer or "please help me upgrade once" spoken by a user, and if any one of the information keywords such as "help you upgrade right soon" or "please help me upgrade once" appears in the target text information, it indicates that the target text information meets the target quality inspection point in the target quality inspection rule, and the appearing keyword is identified in the target text information; if any one of the information keywords of "help you upgrade right soon", "please help me upgrade once" and the like does not appear in the target text information, the target service text data can be directly confirmed as a normal text without errors.

Step S203, acquiring associated text information associated with the target text information from the target service text data, and performing anomaly detection on the associated text information according to an associated quality inspection point in the target quality inspection rule;

specifically, after the target text information hits the target quality inspection point, the associated text information associated with the target text information may be acquired from the target service text data, and the associated text information may be analyzed by using the associated quality inspection point in the target quality inspection rule to determine whether the associated text information hits the associated quality inspection point. The associated text information may refer to the first few words and/or the last few words of the target text information in the target service text data, or may refer to a text having a semantic meaning similar to that of the target text information, the associated quality inspection point may refer to the remaining quality inspection points in the target quality inspection rule depending on the hit condition of the target quality inspection point, and the associated quality inspection point may refer to one or more quality inspection points in the target quality inspection rule. For example, in the service upgrade scenario, 5 sentences before and after the target text information may be used as the associated text information, and the associated quality inspection points in the target quality inspection rule may include two quality inspection points (quality inspection point 1, quality inspection point 2), for example, the quality inspection point 1 may be "you are helped to upgrade successfully" corresponding to the customer service role not being detected, and the quality inspection point 2 may be "please pay attention to check for successful upgrade notification information later" corresponding to the customer service role not being detected.

Step S204, if the associated text information has associated quality inspection points meeting the target quality inspection rule, setting a second identifier corresponding to the associated quality inspection points in the associated text information, and determining target service text data carrying the first identifier and the second identifier as target abnormal text.

Specifically, in the process of performing anomaly detection on the associated text information according to the associated quality control points, if associated quality control points meeting a target quality control rule exist in the associated text information, that is, if the associated text information hits partial quality control points included in the associated quality control points, positions corresponding to the hit partial quality control points in the associated text information may be marked, where the marking information at this time may be referred to as a second identifier, so that an error item may be directly located according to the second identifier when performing manual review later, target service text data carrying the first identifier and the second identifier may be determined as a clue, which may be understood as target service text data suspected of an error, and at this time, the target service text data suspected of an error may be referred to as a first text or a target anomaly text; optionally, if the associated text information hits all the quality inspection points included in the associated quality inspection points, the target service text data may be determined as a detected error text, where the detected error text is the target service text data that is determined by the quality inspection system and does not need to be manually reviewed, and at this time, the determined target service text data may be referred to as a second text or a determination abnormal text. For example, also in a service upgrade scenario, the associated quality inspection points in the target quality inspection rule may include quality inspection point 1: the method comprises the following steps that 'you have helped to upgrade successfully' corresponding to a customer service role is not detected, and a quality inspection point 2: if the associated text message does not detect that the service role corresponding to the notification message of success of later-mentioned attention checking and receiving the upgrade is successful, "help you successfully upgrade" is detected and the service role corresponding to the notification message of success of later-mentioned attention checking and receiving is not detected within a preset time period, it is determined that the upgrade improper rule is hit, and the target service text message can be determined as the second text of the error confirmation (i.e. the abnormal confirmation text); optionally, in the service upgrade scenario as well, if a quality inspection point 1 in the associated quality inspection points is hit in the associated text information: if it is not detected that the service role corresponding to the service role has helped successfully upgrade, and the associated quality inspection point 2 is not hit in the preset time length, that is, the notification information of asking to pay attention to check and receive the upgrade success later is received in the preset time length, it is determined that a part of the quality inspection points hit the improper upgrade rule (that is, the target quality inspection rule), so that the associated information text cannot be accurately judged temporarily according to the target quality inspection rule, therefore, the associated quality inspection point 1 can be marked in the associated text information, and the target service text data carrying the associated quality inspection point 1 is used as the target abnormal text suspected to be abnormal.

The embodiment of the invention determines whether a target quality inspection point exists in the target text information or not according to a target quality inspection rule by acquiring the target text information in the target service text data, sets a first identifier corresponding to the target quality inspection point in the target text information if the target quality inspection point exists, further can acquire associated text information associated with the target text information, determines whether the associated text information has the associated quality inspection point according to the target quality inspection rule, sets a second identifier corresponding to the associated quality inspection point in the associated text information if the associated text information exists, and further can determine the target service text data carrying the first identifier and the second identifier as a target abnormal text. Therefore, in the whole quality inspection process, the fine items with errors can be quickly found out from the target service text data through the obtained target quality inspection rule, the found fine items are marked to obtain the first identification and the second identification, and the target service text data with the marks can be determined as the target abnormal text. In other words, the target abnormal text carries the first identifier and the second identifier, so that the subsequent quality inspection personnel can quickly locate error details according to the marks, the workload of the quality inspection personnel is reduced, and the quality inspection efficiency is improved.

Referring to fig. 3, fig. 3 is a schematic flowchart of another data processing method according to an embodiment of the invention. As shown in fig. 3, the method may include:

step S301, acquiring at least one service voice data, selecting one service voice data from the at least one service voice data as target voice data, and acquiring voice spectrum characteristics in the target voice data;

specifically, the server may obtain all interactive voice data between the customer service and the user from the customer service in multiple channels, that is, all service voice data, may select one service voice data (one complete call between the customer service and the user) from the obtained service voice data as target voice data, and may perform preprocessing on the target voice data, including removing non-voice data in the target voice data (that is, other sounds in the surrounding environment, such as sounds of people walking during the interaction between the customer service and the user), and performing framing processing on the target voice data (a frame length is generally between 10 and 30 milliseconds), and may consider that each frame of voice data is stable, that is, a distribution rule of relevant feature parameters of each frame of voice data is consistent. When extracting the speech frequency spectrum feature from the target speech data after the frame division processing, a feature that can represent the frame speech data can be extracted from each frame speech data, and then a feature sequence of the target speech data, that is, a speech frequency spectrum feature (for example, mel frequency cepstrum coefficient, perceptual linear prediction coefficient, etc.) can be obtained.

Step S302, based on the voice frequency spectrum characteristics and a first recognition model, carrying out voiceprint recognition on the target voice data to obtain a voiceprint recognition result corresponding to the first recognition model; the voiceprint recognition result comprises the matching degree between the voice spectrum characteristic and a plurality of attribute spectrum characteristics in the first recognition model;

specifically, the obtained speech spectrum feature may be input into a first recognition model (i.e., a voiceprint recognition model), and the voiceprint recognition model is trained by using the sample speech data and the attribute spectrum feature corresponding to the sample speech data, that is, has a function of recognizing a speaker. Therefore, the recognition function of the voiceprint recognition model is utilized to obtain a voiceprint recognition result corresponding to the target voice data, and the voiceprint recognition result comprises the matching degree between the voice spectrum feature and the plurality of attribute spectrum features in the voiceprint recognition model. Since each person's voice is different in tone quality, duration, intensity and pitch, different voice spectrum characteristics are presented on the voiceprint atlas, and the voiceprint recognition model has learned characteristics of various human pronunciations (which can be understood as differences between voice spectrum characteristics of different human speeches) from sample voice data. In other words, the above-described voiceprint recognition model has learned what the difference between different persons when uttering the same voice. For example, the sample voice data includes voices of multiple persons, which are respectively a speaker 1, a speaker 2, …, and a speaker n (n is a natural number), and the speaker n can be represented by different tag information, for example, the tag information corresponding to the speaker 1 is a value 1, the tag information corresponding to the speaker 2 is a value 2, …, and the tag information corresponding to the speaker n is a value n, the voiceprint recognition model can learn the relationship between the attribute spectrum feature of each speaker voice and the corresponding tag information through training, and for newly input target voice data, the speaker in the target voice data can be determined by calculating the matching degree between the voice spectrum feature in the target voice data and the attribute spectrum feature corresponding to each of the multiple speakers that have been learned in the voiceprint recognition model.

Step S303, according to the voiceprint recognition result, using the label information associated with the attribute spectrum feature with the highest matching degree with the voice spectrum feature as a first role recognized from the target voice data;

specifically, according to the matching degree between the speech spectrum feature in the target speech data and the plurality of attribute spectrum features in the voiceprint recognition model, the tag information corresponding to the attribute spectrum feature with the highest matching degree may be determined as the first role (i.e., the customer service role) in the target speech data. In the field of quality inspection, the voice spectrum features in the target voice data are matched with the attribute spectrum features of all recorded customer service voices in the voiceprint recognition model to determine the customer service roles in the target voice data, and in addition, the voice spectrum features can be matched with the recorded customer service common techniques (such as 'good will for your service') while being matched with the attribute spectrum features of the recorded customer service voices, so that the customer service roles can be determined more quickly. It should be noted that, in the field of quality inspection, each voice data only contains voices of a client role and a user role, so that when performing voiceprint recognition on target voice data, only a part of voices belonging to a customer service role in the target voice data need to be distinguished, voices belonging to the user role can be determined, and it is not necessary to necessarily recognize specific customer service personnel in the target voice data, so that when detecting a customer service common speech technology in the target voice data, the voiceprint recognition model can recognize the voice corresponding to the common speech technology to determine the customer service role; when the customer service common dialogs are not detected, the customer service role can be determined by matching with the attribute spectrum characteristics of all the customer service voices which are recorded.

Step S304, determining the rest roles except the first role in the target voice data as second roles, and converting the target voice data carrying the first role and the second role into service text data based on a control frequency parameter and a second identification model;

specifically, after the customer service role in the target voice data is determined, another role in the target voice data may be determined as a second role (i.e., a user role), a control frequency parameter may be obtained, and the target voice data that has been subjected to the customer service role and the user role differentiation is input into a second recognition model (i.e., a voice recognition model) based on the control frequency parameter, where the voice recognition model has been trained through the sample voice data in the corpus database and the text information corresponding to the sample voice data, i.e., has a text conversion function. The target voice data can be converted into service text data using the voice recognition model. The control frequency parameter may be used to control the amount of service voice data converted into service text data in a unit time, that is, to indicate that the acquired service voice data is converted into service text data in batches.

Step S305, when each service voice data in the at least one service voice data is taken as a target voice data, obtaining service text data corresponding to each service voice data, and selecting one service text data from the converted service text data as a target service text data;

specifically, when all the obtained service voice data are used as target voice data, the voiceprint recognition model may be used to determine a customer service role and a user role in each service voice data, convert each service voice data subjected to role differentiation into corresponding service text data, and select one service text data from all the service text data obtained through conversion as the target service text data. It can be understood that, when the service voice data is converted into the service text data, the voice recognition model may perform parallel processing, that is, a plurality of target voice data may be simultaneously input into the voice recognition model to obtain service text data corresponding to the plurality of target voice data, and the specific target voice amount may be determined by the obtained control frequency parameter.

Step S306, counting the target quantity of all the service text data obtained by conversion;

specifically, when all the service voice data are converted into service text data, the server may count the total amount (i.e., the target amount) of all the service text data, that is, the number of times of call between the obtained customer service and the user.

Step S307, if the target number is detected to be greater than the backup number threshold, performing backup processing on the service text data exceeding the backup number threshold in all the service text data;

specifically, a backup quantity threshold for storing service text data may be obtained, and if it is detected that the total quantity of the service text data is greater than the backup quantity threshold, backup processing may be performed on the service text data exceeding the backup quantity threshold in all the service text data, that is, when the total quantity of the service text data exceeds the storage space of the storage database, the exceeding service text data may be transferred to the backup storage database for backup processing. For example, if the total amount of all service text data is counted to be 70 ten thousand, and the storage space of the storage database (i.e., the backup number threshold) is 50 ten thousand, then the exceeding 20 ten thousand service text data can be transferred to the backup storage database.

Step S308, acquiring first text data corresponding to the first role and second text data corresponding to the second role from the target service text data;

specifically, after one service text data is selected from all the service text data obtained through conversion as the target service text data, the first text data corresponding to the first role, that is, the speaking content corresponding to the customer service role, and the second text data corresponding to the second role, that is, the speaking content corresponding to the user role, can be obtained from the target service text data. Optionally, the first text data is not limited to all the utterances of the customer service role in the target service text data, and may be a sentence spoken by the customer service role, or several sentences spoken by the customer service role, which may be determined according to a specific scenario, and is not limited herein. Similarly, the second text data is not specifically limited, and may be determined according to a specific scenario.

Step S309, obtaining a target quality inspection rule from a plurality of quality inspection rules, taking the first text data and/or the second text data as target text information, selecting a quality inspection point with the highest priority from a plurality of quality inspection points contained in the target quality inspection rule as a target quality inspection point, and performing anomaly detection on the target text information based on the target quality inspection point;

specifically, the first text data and/or the second text data may be used as the target text information. In other words, the target text data may be text content corresponding to the customer service role, may also be text content corresponding to the user role, and may also be a segment of continuous session content between the customer service role and the user role. The target quality inspection rule matched with the scene to which the target service text data belongs can be obtained from the plurality of quality inspection rules in the intelligent analysis engine, the priority of the plurality of quality inspection points in the target quality inspection rule is obtained, the quality inspection point with the highest priority is used as the target quality inspection point, and the target text information can be subjected to anomaly detection based on the target quality inspection point. The target quality inspection rule is a quality inspection rule set for a scene to which the target service text message belongs, the target quality inspection rule may include a plurality of quality inspection points, and if one quality inspection point exists in all the quality inspection points of the target quality inspection rule, and the hit conditions of the remaining quality inspection points depend on the context after the hit of the quality inspection point, the quality inspection point may be considered to have the highest priority among all the quality inspection points included in the target quality inspection rule, and the quality inspection point may be used as the target quality inspection point, which may also be referred to as a pre-quality inspection point. For example, in a scenario of a customer service hold line (referring to a user who has not responded for a certain time after customer service), a target quality inspection rule is used to detect service text information of a user who has not responded for a long time due to customer service itself, a target quality inspection point in the target quality inspection rule may be a session interval between a customer service role and a user role for more than a certain time (e.g., 50 s), hits of the remaining quality inspection points in the target quality inspection rule all depend on hits of the target quality inspection point, that is, when the target text information hits the target quality inspection point, the remaining quality inspection points may be determined continuously from a hit context of the target quality inspection point, and when the target text information misses the target quality inspection point, the target service text data may be determined to miss the remaining quality inspection points directly.

Step S310, if a target quality inspection point meeting a target quality inspection rule exists in the target text information, setting a first identification corresponding to the target quality inspection point in the target text;

for a specific implementation manner of the step S310, reference may be made to the description of the step S202 in the embodiment corresponding to fig. 2, and details are not described here again.

Step S311, using the target quality inspection point corresponding to the first identifier as a starting inspection point, and acquiring a detection range threshold corresponding to the target quality inspection rule;

specifically, after the target text information hits the target quality inspection point, the target quality inspection point may be used as a start inspection point, and a detection range threshold set for the target quality inspection point in the target quality inspection rule is obtained, where the detection range threshold may include a first detection range threshold and/or a second detection range threshold, where the first detection range threshold may be a range threshold for starting forward detection with the start inspection point, and the second detection range threshold may be a range threshold for starting backward detection with the start inspection point.

Step S312, determining the text information between the starting check point and the detection range threshold value as the associated text information associated with the target text information;

specifically, when the detection range threshold includes a first detection range threshold, the text information between the first detection range thresholds before the start checkpoint may be determined as the associated text information associated with the target text information; when the detection range threshold includes a second detection range threshold, text information between the second detection range thresholds after the starting checkpoint may be determined as associated text information associated with the target text information; when the detection range threshold includes a first detection range threshold and a second detection range threshold, text information between the first detection range threshold before the start checkpoint and text information between the second detection range threshold after the start checkpoint may be determined as the associated text information. For example, if the start check point is the 10 th sentence in the target service text data, the first detection range threshold is 3 sentences, and the second detection range threshold is 5 sentences, if the detection range threshold includes the first detection range threshold, the associated text information is the 7 th sentence to the 10 th sentence in the target service text data, if the detection range threshold includes the second detection range threshold, the associated text information is the 10 th sentence to the 15 th sentence in the target service text data, and if the detection range threshold includes the first detection range threshold and the second detection range threshold, the associated text information is the 7 th sentence to the 15 th sentence in the target service text data.

Step S313, obtaining the associated quality inspection points in the target quality inspection rule, wherein the associated quality inspection points have an association relation with the target quality inspection points, and carrying out abnormal detection on the associated text information according to the associated quality inspection points;

specifically, after determining the associated text information associated with the target text information, the method may acquire associated quality inspection points in the target quality inspection rule, that is, the remaining quality inspection points depending on the hit condition of the target quality inspection point, perform abnormality detection on the associated text information according to the associated quality inspection points, and determine the hit condition of the associated quality inspection points in the associated text information. For example, in a customer service hold line scenario, the associated quality inspection point may include a condition that the user actively waits for the customer service, such as "i find a bank card" or the like, which is said by the user, and if a sentence similar to "i find a bank card" is not detected in the associated text information, it indicates that the relevant text hits the associated quality inspection point, otherwise, the associated quality inspection point is not hit.

Step S314, if the associated text information has the associated quality inspection point meeting the target quality inspection rule, setting a second identifier corresponding to the associated quality inspection point in the associated text information, and determining the target service text data carrying the first identifier and the second identifier as a target abnormal text.

For a specific implementation manner of the step S314, reference may be made to the description of the step S204 in the embodiment corresponding to fig. 2, which is not described herein again.

For example, taking a quality inspection case in a customer service hold line scene as an example, in order to accurately check a service work order (i.e. the service voice data) of the hold line, for a service work order, the specific implementation manner is as follows:

dividing voice in the service work order into two roles, namely role 1and role 2, by a voiceprint recognition technology, matching with the recorded customer service voiceprint and the common speech technique of customer service to determine the customer service role in the service work order, converting voice data in the service work order into text number by a voice-to-text technology, and marking the customer service role in the text data, so that the other role in the service work order can be determined as a user role; checking that the conversation interval between the customer service role and the user role in the service work order exceeds 50 seconds, and the latter sentence is the conversation of the customer service speaking, and the mark is checkpoint 1= true, if the miss mark is checkpoint 1= false; when the check point 1 is not larger than the future, the above 5 sentences of the check point 1 are searched, and the condition that a user actively requires customer service waiting is eliminated, such as 'i find a bank card', and the result is recorded as the result of the check point 2; when the check point 1 is not larger than the threshold, the context 3 sentence of the check point 1 is searched, and the condition that the user is waiting in the customer service statement is excluded, such as 'you input once', 'have a core and do', and the condition is recorded as the result of the check point 3; when the check point 1 is not = true, searching the above 3 sentences of the check point 1, and excluding the condition that the user inputs Interactive Voice Response (IVR) key again in the customer service utterance, if 'please input according to Voice prompt', recording the result as the check point 4; when the check point 1 is not = true, the above 3 sentences of the check point 1 are searched, and the condition that the user waits for customer service to dial another number for verification in customer service speaking is eliminated, such as "i call your another number for verification in the past, ask you to listen to a piece of music", and the result is recorded as the result of the check point 5; finally, the inspection results need to be combined, and the combination formula is as follows: 1and (not (2 or3or4or 5)), and 1-5 refer to the hit of each check point, and if the result is obtained by performing Boolean operation, if not and (not or false), then the rule of hitting the customer service hold line is regarded as detected, and the detected rule indicates that the customer service order is confirmed to have an error. Wherein, check point 1 is a target quality inspection point, check point 1= true indicates a hit target quality inspection point, check point 1= false indicates a miss target quality inspection point, check point 2-check point 5 is a correlated quality inspection point, the above-mentioned above 3 sentences or above-mentioned 5 sentences are the above-mentioned first detection range threshold, below-mentioned above-mentioned 3 sentences are the second detection range threshold, and the above-mentioned combination formula 1and (not (2 or3or4or 5)) are the target quality inspection rule in the customer service hold line scene. When the service work order hits the target quality inspection rule, the service work order can be confirmed to be a wrong service work order; when the service work order hits in the checking point 1and the target quality inspection rule is not hit, the service work order can be used as a suspected error service work order, namely, the service work order is determined as a target abnormal text.

Referring to fig. 4, fig. 4 is a schematic diagram of a framework structure of a customer service quality inspection method according to an embodiment of the present invention. As shown in the figure, in the customer service quality inspection process, the quality inspection session analysis service 43 in the intelligent analysis engine may obtain service voice data from the call center 41, and at the same time, may perform a timed backup on the service voice data in the call center 41, and store all the service voice data in the data backup server 42, so as to query the original voice data between the customer and the user in the following. After obtaining the service voice data from the call center 41, the quality inspection session analysis service 43 may perform voice recognition 45 on the obtained service voice data based on the control frequency 431 parameter, convert the service voice data into service text data, and perform voiceprint recognition on the service text data in the process of performing the voice recognition 45, that is, perform role determination 434 on the service voice data, where the specific role determination 434 process and the voice recognition 45 process may refer to step S301 to step S305 in the embodiment corresponding to fig. 3, and are not described herein again. After the service voice data is converted into the text data, the converted service text data may be stored in the voice-to-text main server 46, when the total amount of the service text data is too large, and the voice-to-text main server 46 cannot store the service text data, the excessive service text data may be stored in the voice-to-text backup server 47, the quality inspection session analysis service 43 may obtain the service text data in the voice-to-text main server 46 and the voice-to-text backup server 47, and analyze the obtained service text data, where the algorithm analysis 433 may represent obtaining a target quality inspection rule corresponding to the service text data, including performing intelligent identification on associated quality inspection points with semantic matching, and the detection exception 432 may represent performing exception detection on the service text data according to the target quality inspection rule, and a specific detection process may refer to steps S308 to S314 in the embodiment corresponding to fig. 3, which is not described herein again. The quality inspection result generated by analyzing the service text data by the quality inspection session analysis service 43 and the service text data with the identification information set in the quality inspection process can be stored in the data warehouse 44, and meanwhile, the quality inspection session analysis service 43 can also acquire the service text data from the data warehouse 44 to train so as to update the quality inspection rule.

In the embodiment of the invention, service text data corresponding to the service voice data can be obtained by performing voiceprint recognition and voice recognition on the obtained service voice data, first text data corresponding to a customer service role and second text data corresponding to a user role can be obtained from the service text data, then target text data can be determined according to the first text data and the second text data, whether a target quality inspection point exists in the target text information or not is determined according to a target quality inspection rule, if the target quality inspection point exists, a first identifier corresponding to the target quality inspection point is set in the target text information, then associated text information associated with the target text information can be obtained, whether an associated quality inspection point exists in the associated text information or not is determined according to the target quality inspection rule, if the associated quality inspection point exists, a second identifier corresponding to the associated quality inspection point is set in the associated text information, and then the target service text data carrying the first identifier and the second identifier can be determined as a target abnormal text. Therefore, in the whole quality inspection process, by means of the obtained target quality inspection rule and different roles in the target service text data, the detail items with errors can be quickly found out from the target service text data, the found detail items are marked to obtain the first identification and the second identification, and the target service text data with the marks can be determined as the target abnormal text. In other words, the target abnormal text carries the first identifier and the second identifier, so that the subsequent quality inspection personnel can quickly locate error details according to the marks, the workload of the quality inspection personnel is reduced, and the quality inspection efficiency is improved.

Referring to fig. 5, fig. 5 is a schematic flowchart of another data processing method according to an embodiment of the present invention. As shown in fig. 5, the method may include:

step S501, obtaining at least one service voice data, respectively converting each service voice data in the at least one service voice data into service text data, and selecting one service text data from the converted service text data as a target service text data;

for a specific implementation manner of the step S501, reference may be made to the description of the step S301 to the step S305 in the embodiment corresponding to fig. 3, which is not described herein again.

Step S502, acquiring target service text data, acquiring target text information from the target service text data, and performing anomaly detection on the target text information according to a target quality inspection point in a target quality inspection rule;

step S503, if a target quality inspection point meeting a target quality inspection rule exists in the target text information, setting a first identifier corresponding to the target quality inspection point in the target text;

step S504, obtaining the associated text information associated with the target text information from the target service text data, and carrying out abnormal detection on the associated text information according to the associated quality inspection point in the target quality inspection rule;

step S505, if the associated text information has associated quality inspection points meeting the target quality inspection rule, setting a second identifier corresponding to the associated quality inspection points in the associated text information, and determining target service text data carrying the first identifier and the second identifier as a target abnormal text;

for a specific implementation manner of the step S502 to the step S505, reference may be made to the description of the step S201 to the step S204 in the embodiment corresponding to fig. 2, and details are not repeated here.

Step S506, adding the target abnormal texts to an abnormal text data set, and sending all the target abnormal texts in the abnormal text data set to a quality inspection terminal;

specifically, after the target abnormal text is determined, the target abnormal text may be added to an abnormal text data set, where the abnormal text data set includes all abnormal texts in the quality inspection process, that is, all service text data suspected of being erroneous, and all target abnormal texts in the abnormal text data set are sent to a quality inspection terminal, that is, to a quality inspector console. After receiving the target abnormal text, the quality inspection terminal can randomly extract a part of target abnormal text from all the received target abnormal texts for manual rechecking, and the quality inspection process mainly comprises the following steps:

finding error types (such as service etiquette types, misoperation types and the like) needing to be detected by adopting quality inspection rules in daily quality inspection in a manual mode; further analyzing the observation points in the manual quality inspection, and sorting the observation points into check points, namely sorting several key points of main inspection in the manual quality inspection into check points; configuring each check point and context dependency relationship between the check points, that is, determining an association relationship between the check points, which may be n sentences above, n sentences below, n sentences above and so on (n is a natural number); combining check points according to the calculation conditions of the OR and non-AND priority operators, namely determining a quality inspection rule so as to achieve the purpose of quality inspection; the hit condition of the quality inspection rule can be instantly verified by selecting one target abnormal text from all the obtained target abnormal texts, for example, a configuration page of a quality inspection terminal has a quick verification function, and after one target abnormal text is input, the hit condition of the rule and the hit condition of the inspection point can be returned; and according to the hit condition, the final quality inspection result of the target abnormal text can be confirmed, namely service text data suspected to be wrong is given to a quality inspection rule in an intelligent analysis engine, and whether the service text data is wrong can be confirmed after manual rechecking. The quality inspection terminal can send the quality inspection result after the target abnormal text is manually rechecked and the corresponding quality inspection rule to the intelligent analysis engine.

Step S507, obtaining abnormal quality inspection points determined by the quality inspection terminal based on all target abnormal texts, and updating the target quality inspection rules based on the abnormal quality inspection points;

specifically, after receiving the quality inspection result obtained after the target abnormal text is manually rechecked and the corresponding quality inspection rule, the abnormal quality inspection point in the manual quality inspection rule, namely the newly added inspection point, can be obtained according to the rule hit condition in the quality inspection result, and the abnormal quality inspection point is added to the target quality inspection rule, so as to update the target quality inspection rule. For example, in a backtracking scenario, a target quality inspection point in the target quality inspection rule is a "backtracking" keyword appearing in the speech content corresponding to the user, and an associated quality inspection point in the target quality inspection point may be a word for excluding a customer service for soothing, for example: the method comprises the steps of ' thank you for supporting the game all the time ', ' hope you can consider whether to determine whether to quit, when a soothing language similar to the associated quality inspection point appears in target service text data, ' ask what reason is why the ask is, what can help you, and the target quality inspection rule cannot judge the ask, so that the target service text data is determined as a target abnormal text, and after manual review, quality inspection shows that the target service text data has no error and belongs to a normal text, and the ' ask why the ask is because reason is, what can help you ' is ' can be added to the target quality inspection rule as a new check point according to a quality inspection result of the manual review.

Step S508, adding the target abnormal text to an abnormal text data set, obtaining a historical abnormal text corresponding to the abnormal text data set, and obtaining a training sample set based on the historical abnormal text;

specifically, after the target abnormal text is added to the abnormal text data set, all historical abnormal texts in the abnormal text data set can be obtained, and all the obtained historical abnormal texts are used as training samples. The historical abnormal text refers to a target abnormal text confirmed in all quality inspection processes before the quality inspection.

Step S509, training a quality inspection model corresponding to the target quality inspection rule based on the training sample set, and updating the quality inspection model; and the abnormal quality inspection points carried in the updated quality inspection model are used for updating the target quality inspection rule.

Specifically, based on the historical abnormal texts in the training sample set, a quality inspection model corresponding to the target quality inspection rule can be trained, each historical abnormal text can be used as a sample data to train the quality inspection model, and after the training is completed, the quality inspection model can obtain better model parameters. In other words, the trained quality inspection model replaces the quality inspection model corresponding to the target quality inspection rule, and the updating process of the quality inspection model is completed. The updated quality inspection model carries abnormal quality inspection points, and the target rule can be updated according to the abnormal quality inspection points. The quality inspection model can be a convolutional neural network, a deep neural network, a generative countermeasure network, and the like. All suspected wrong service text data (namely target abnormal texts) subjected to quality detection by the target quality detection rule are summarized, and deep learning is performed on the suspected wrong service text data, namely, the characteristics in the suspected wrong service text data are independently learned by a quality detection model, so that better model parameters are obtained, the target quality detection rule is optimized, and the quality detection quality is improved. For example, there is a mechanism that can recognize a red ball and a dark blue ball, but cannot recognize a light blue ball, in a box, with balls of three colors of red, dark blue, and light blue, and the mechanism can be optimized by learning the characteristics of the light blue ball, so that the updated mechanism can recognize the light blue ball, for example, the identification mechanism that updates the identification manner for the dark blue ball in the original mechanism to a blue ball (including the dark blue ball and the light blue ball). If the mechanism represents the target quality inspection rule, the red ball can represent the text data which is confirmed to have no error in the service text data, the dark blue ball can represent the text data which is confirmed to have an error in the service text data, the light blue ball can represent the text data which is suspected to have an error in the service text data, the characteristic of learning the light blue ball can be represented by learning the text data which is suspected to have an error by using a quality inspection model, and the updated mechanism can represent the updated target quality inspection rule.

It should be noted that, the quality inspection result obtained through manual review described in the above steps S506 to S507 updates the target quality inspection rule, and the historical abnormal text obtained through deep learning described in the steps S508 to S509 continuously optimizes the target quality inspection rule into two independent update modes of the target quality inspection rule, and the update modes are not interfered with each other. In other words, the two target quality inspection rule updating modes can be performed successively or simultaneously.

Please refer to fig. 6 a-6 b, which are schematic structural diagrams of a framework of another customer service quality inspection method according to an embodiment of the present invention. As shown in fig. 6a, the frame may include: interaction record data 602 is acquired from each channel customer service 601, and data preprocessing 603 is performed on the acquired interaction record data 602, where the preprocessing includes converting acquired voice 6031 data into text 6033 data through an Automatic Speech Recognition technology 6032 (ASR), and for a specific process of converting voice into text, reference may be made to step S301 to step S305 in the embodiment corresponding to fig. 3, and details are not described here again. The converted text data is input to the quality inspection library 604, and in the quality inspection library 604, the intelligent analysis engine 6041 may analyze the converted text data to obtain a quality inspection result corresponding to each text data, and a specific quality inspection process may refer to steps S308 to S314 in the embodiment corresponding to fig. 3, which is not described herein again. For suspected erroneous text data detected by the quality inspection of the intelligent analysis engine 6041, manual review may be performed through the quality inspector console 6042, the quality inspection rule in the intelligent analysis engine 6041 may be updated according to the quality inspection result after the manual review, that is, the quality inspection rule in the intelligent analysis engine 6041 is complemented, the quality inspector console 6042 is the quality inspection terminal in the embodiment corresponding to fig. 3, and for the specific implementation process of the manual review and the update of the target quality inspection rule, reference may be made to steps S506 to S509 in the embodiment corresponding to fig. 5, which is not described herein again. The operational flow data 605 is primarily used to control the flow of data between the interactive log data 602 to the quality testing process in the quality testing library 604.

Optionally, the interaction record data 602 acquired from each channel customer service 601 may also include text data, for example, data acquired from an online customer service, and if the acquired interaction record data 602 is text data, the data preprocessing 603 need not be executed, and the data may be directly input into a quality inspection library.

As shown in fig. 6B, the framework may represent a specific quality inspection process, the quality inspection rule 606 in the quality inspection process may verify the rule in the session analysis model 607 by using a combination of a check point, an applicable range 6065 and a logical relationship 6066 (e.g., and/or not), where the check point is a minimum unit constituting the quality inspection rule 606, and may be a minimum unit of any service order content, including but not limited to a keyword 6061, a service order attribute 6062, a dialog interval 6063, and collation information used in all manual quality inspections such as semantic matching 6064, and the rule in the session analysis model 607 may include a plurality of rules for checking problem service orders under various scenarios, for example, rule a 6071 may represent a hold line rule, rule B6072 may represent a fallback rule, and rule C6073 may represent a recommended product rule. The session analysis model 607 can perform quality inspection on the service work orders in the quality inspection library 608 based on the above rules, and can obtain a cue sheet (i.e. a service work order suspected of error detected by the intelligent quality inspection is also equal to the abnormal text mentioned in the above embodiments), the Artificial Intelligence (AI) cue 609 includes a plurality of cues, that is, a plurality of cue sheets can be detected, such as cue 1 6091, cue 2 6092, and cue 3 6093, it is to be noted that rule a 6071, rule B6072, rule C6073 and cue 1 6091, cue 2 6092, and cue 3 6093 are not in a one-to-one correspondence relationship, for the same rule, a plurality of cue sheets can be detected, and for the cue sheet detected by the session analysis model 607, the manual quality inspection bench can be transmitted to perform manual review 610, and the quality inspection rule 606 can be updated according to the quality inspection result, the rule hit condition, and the hit condition of the inspection point, for example, a check point is added. The quality control rule 606 can also be optimized through the machine learning commonality difference 611, that is, the machine learning is used to learn the clue list, and the quality control rule 606 is intelligently optimized.

Referring to fig. 7, fig. 7 is a timing diagram illustrating a customer service quality inspection method according to an embodiment of the present invention. As shown, the method may include:

step S701, the front end sends the work order content to the background;

specifically, after acquiring the work order content (i.e., the service work order content) from the customer service of each channel, the front end may send the acquired service work order content to the back end.

Step S702, checking work order information by a background;

specifically, after receiving the content of the service work order sent by the foreground, the background can detect the data type of the service work order, and if the service work order is voice data, the service work order is converted into text data, and then the service work order information is checked; if the service work order is text data, the service work order information can be directly checked. The checked service work order information mainly comprises service work order remarks, service work order labels, conversation contents in the service work orders and the like.

Step S703, extracting rules;

specifically, the quality inspection rule matched with the service work order information can be extracted according to the service work order information. For example, for service work order information of a product recommendation class, a product recommendation rule needs to be extracted; for the upgrade-type service work order information, an upgrade rule needs to be extracted.

Step S704, the front end transmits the rule to the background;

specifically, the front end can transmit the extracted quality inspection rule to the background, so that the background can perform quality inspection on the service work order according to the quality inspection rule.

Step S705, checking logic by background check;

specifically, after receiving the quality inspection rule sent by the background, the background can check the logical relationship of each inspection point in the quality inspection rule, and further determine the hit condition of each inspection point and the hit condition of the quality inspection rule, the service work order which does not hit the quality inspection rule can be determined as having no error, the service work order which hits the quality inspection rule can be determined as having an error, and the rule which is still in adjustment can be used as the service work order which is suspected to have an error, namely a quality inspection clue.

Step S706, the background sends quality inspection clues to the manual work;

specifically, after the background intelligently performs quality inspection on the result of the service work order, the background can send a quality inspection clue to the manual quality inspection operation console, that is, the service work order suspected of making an error is sent to the manual quality inspection operation console, so that the quality inspection personnel perform manual rechecking.

Step S707, sending the quality inspection result to the background manually;

specifically, the manual quality inspection console can transmit the quality inspection result after manual recheck back to the background, and the background can analyze the quality inspection result after manual recheck.

Step S708, adjusting rules;

specifically, the adjustment information of the quality inspection rule may be obtained according to the quality inspection result after manual review, for example, adding a check point or optimizing a check point.

Step S709, the front end updates the rule.

Specifically, the front end may update the quality control rule according to the adjustment information to obtain an improved scheme or a new rule of the quality control rule.

The front end and the background correspond to the server 100 in the embodiment corresponding to fig. 1, and the manual operation corresponds to the quality inspection terminal 200 in the embodiment corresponding to fig. 1.

The embodiment of the invention determines whether a target quality inspection point exists in the target text information or not according to a target quality inspection rule by acquiring the target text information in the target service text data, sets a first identifier corresponding to the target quality inspection point in the target text information if the target quality inspection point exists, further can acquire associated text information associated with the target text information, determines whether the associated text information has the associated quality inspection point according to the target quality inspection rule, sets a second identifier corresponding to the associated quality inspection point in the associated text information if the associated text information exists, and further can determine the target service text data carrying the first identifier and the second identifier as a target abnormal text. Therefore, in the whole quality inspection process, the fine items with errors can be quickly found out from the target service text data through the obtained target quality inspection rule, the found fine items are marked to obtain the first identification and the second identification, and the target service text data with the marks can be determined as the target abnormal text. In other words, the target abnormal text carries the first identifier and the second identifier, so that subsequent quality inspection personnel can be helped to quickly locate error details according to the marks, the workload of the quality inspection personnel is reduced, the quality inspection efficiency is improved, and the target quality inspection rule can be updated through the abnormal quality inspection points in manual review or the abnormal quality inspection points obtained by deep learning of historical abnormal texts, so that the quality inspection quality can be improved.

Referring to fig. 8, fig. 8 is a schematic structural diagram of a data processing apparatus according to an embodiment of the present invention. As shown in fig. 8, the data processing apparatus 1 may include: a text information acquisition module 101, a first setting module 102, an associated information acquisition module 103, and a second setting module 104;

the text information acquisition module 101 is configured to acquire target service text data, acquire target text information from the target service text data, and perform anomaly detection on the target text information according to a target quality inspection point in a target quality inspection rule;

a first setting module 102, configured to set, if a target quality inspection point meeting a target quality inspection rule exists in the target text information, a first identifier corresponding to the target quality inspection point in the target text;

the associated information acquisition module 103 is configured to acquire associated text information associated with the target text information from the target service text data, and perform anomaly detection on the associated text information according to an associated quality inspection point in the target quality inspection rule;

a second setting module 104, configured to set a second identifier corresponding to the associated quality inspection point in the associated text information if the associated quality inspection point in the target quality inspection rule is met in the associated text information, and determine target service text data carrying the first identifier and the second identifier as a target abnormal text.

For specific functional implementation manners of the text information obtaining module 101, the first setting module 102, the associated information obtaining module 103, and the second setting module 104, reference may be made to steps S201 to S204 in the embodiment corresponding to fig. 2, which is not described herein again.

As shown in fig. 8, the data processing apparatus 1 may further include: the system comprises a text conversion module 105, a statistical module 106, a backup module 107, an abnormal text sending module 108, a first updating module 109, a training sample determining module 110 and a second updating module 111;

the text conversion module 105 is configured to obtain at least one service voice data, convert each service voice data in the at least one service voice data into service text data, and select one service text data from the converted service text data as a target service text data;

a statistical module 106, configured to count target quantities of all service text data obtained through conversion;

a backup module 107, configured to perform backup processing on service text data exceeding the backup number threshold in all the service text data if it is detected that the target number is greater than the backup number threshold;

an abnormal text sending module 108, configured to add the target abnormal text to an abnormal text data set, and send all target abnormal texts in the abnormal text data set to a quality inspection terminal;

a first updating module 109, configured to obtain an abnormal quality inspection point determined by the quality inspection terminal based on all target abnormal texts, and update the target quality inspection rule based on the abnormal quality inspection point;

a training sample determining module 110, configured to add the target abnormal text to an abnormal text data set, obtain a historical abnormal text corresponding to the abnormal text data set, and obtain a training sample set based on the historical abnormal text;

a second updating module 111, configured to train, based on the training sample set, a quality inspection model corresponding to the target quality inspection rule, and update the quality inspection model; and the abnormal quality inspection points carried in the updated quality inspection model are used for updating the target quality inspection rule.

The specific functional implementation manners of the text conversion module 105, the statistical module 106, and the backup module 107 may refer to step S301 to step S307 in the embodiment corresponding to fig. 3, which are not described herein again, and the specific functional implementation manners of the abnormal text sending module 108, the first updating module 109, the training sample determining module 110, and the second updating module 111 may refer to step S506 to step S509 in the embodiment corresponding to fig. 5, which is not described herein again.

As shown in fig. 8, the text information obtaining module 101 may include: a first acquisition unit 1011, a target quality inspection point determination unit 1012;

a first obtaining unit 1011, configured to obtain first text data corresponding to the first role and second text data corresponding to the second role from the target service text data;

the target quality inspection point determining unit 1012 is configured to obtain a target quality inspection rule from a plurality of quality inspection rules, use the first text data and/or the second text data as target text information, select a quality inspection point with the highest priority from a plurality of quality inspection points included in the target quality inspection rule as a target quality inspection point, and perform abnormality detection on the target text information based on the target quality inspection point.

The specific functional implementation manners of the first obtaining unit 1011 and the target quality inspection point determining unit 1012 may refer to steps S308 to S309 in the embodiment corresponding to fig. 3, which is not described herein again.

As shown in fig. 8, the association information obtaining module 103 may include: a check point determination unit 1031, a first determination unit 1032, an abnormality detection unit 1033;

a check point determining unit 1031, configured to use the target quality inspection point corresponding to the first identifier as a starting check point, and obtain a detection range threshold corresponding to the target quality inspection rule;

a first determination unit 1032 configured to determine text information between the start checkpoint and the detection range threshold as associated text information associated with the target text information;

an anomaly detection unit 1033, configured to obtain an associated quality inspection point in the target quality inspection rule, where the associated quality inspection point has an association relationship with the target quality inspection point, and perform anomaly detection on the associated text information according to the associated quality inspection point.

The specific functional implementation manners of the checkpoint determining unit 1031, the first determining unit 1032, and the anomaly detecting unit 1033 may refer to step S311 to step S313 in the embodiment corresponding to fig. 3, which is not described herein again.

As shown in fig. 8, the text conversion module 105 may include: a selection unit 1051, a voiceprint recognition unit 1052, a first role determination unit 1053, a second role determination unit 1054, a text data determination unit 1055;

a selecting unit 1051, configured to select one service voice data from the at least one service voice data as target voice data, and obtain a voice spectrum feature in the target voice data;

a voiceprint recognition unit 1052, configured to perform voiceprint recognition on the target voice data based on the voice spectrum feature and the first recognition model, so as to obtain a voiceprint recognition result corresponding to the first recognition model; the voiceprint recognition result comprises the matching degree between the voice spectrum characteristic and a plurality of attribute spectrum characteristics in the first recognition model;

a first role determination unit 1053, configured to use, according to the voiceprint recognition result, tag information associated with an attribute spectrum feature having a highest matching degree with the voice spectrum feature as a first role recognized from the target voice data;

a second role determination unit 1054, which determines the remaining roles except the first role in the target voice data as second roles, and converts the target voice data carrying the first role and the second role into service text data based on a control frequency parameter and a second recognition model;

a text data determining unit 1055, configured to obtain service text data corresponding to each service voice data when each service voice data in the at least one service voice data is taken as target voice data.

For specific functional implementation manners of the selecting unit 1051, the voiceprint identifying unit 1052, the first role determining unit 1053, the second role determining unit 1054, and the text data determining unit 1055, reference may be made to steps S301 to S305 in the embodiment corresponding to fig. 3, which is not described herein again.

The embodiment of the invention determines whether a target quality inspection point exists in the target text information or not according to a target quality inspection rule by acquiring the target text information in the target service text data, sets a first identifier corresponding to the target quality inspection point in the target text information if the target quality inspection point exists, further can acquire associated text information associated with the target text information, determines whether the associated text information has the associated quality inspection point according to the target quality inspection rule, sets a second identifier corresponding to the associated quality inspection point in the associated text information if the associated text information exists, and further can determine the target service text data carrying the first identifier and the second identifier as a target abnormal text. Therefore, in the whole quality inspection process, the fine items with errors can be quickly found out from the target service text data through the obtained target quality inspection rules, the found fine items are marked to obtain the first identification and the second identification, and the target service text data with the marks can be determined as the target abnormal text. In other words, the target abnormal text carries the first identifier and the second identifier, so that subsequent quality inspection personnel can be helped to quickly locate error details according to the marks, the workload of the quality inspection personnel is reduced, the quality inspection efficiency is improved, and the target quality inspection rule can be updated through the abnormal quality inspection points in manual review or the abnormal quality inspection points obtained by deep learning of historical abnormal texts, so that the quality inspection quality can be improved.

Referring to fig. 9, fig. 9 is a schematic structural diagram of another data processing apparatus according to an embodiment of the present invention. As shown in fig. 9, the data processing apparatus 1000 may correspond to the server 100 in the embodiment corresponding to fig. 1, and the data processing apparatus 1000 may include: the processor 1001, the network interface 1004, and the memory 1005, and the data processing apparatus 1000 may further include: a user interface 1003, and at least one communication bus 1002. Wherein a communication bus 1002 is used to enable connective communication between these components. The user interface 1003 may include a Display screen (Display) and a Keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a standard wireless interface. The network interface 1004 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface). The memory 1004 may be a high-speed RAM memory or a non-volatile memory (e.g., at least one disk memory). The memory 1005 may alternatively be at least one memory device located remotely from the processor 1001. As shown in fig. 9, a memory 1005, which is a kind of computer storage medium, may include therein an operating system, a network communication module, a user interface module, and a device control application program.

In the data processing apparatus 1000 shown in fig. 9, the network interface 1004 may provide a network communication function; the user interface 1003 is an interface for providing a user with input; and the processor 1001 may be used to invoke a device control application stored in the memory 1005 to implement:

In one embodiment, the processor 1001 may further implement:

the method comprises the steps of obtaining at least one service voice datum, respectively converting each service voice datum in the at least one service voice datum into service text data, and selecting one service text datum from the converted service text data as a target service text datum.

In one embodiment, the processor 1001 may further implement:

counting the target quantity of all service text data obtained by conversion;

if the target number is detected to be larger than the backup number threshold value, performing backup processing on the service text data exceeding the backup number threshold value in all the service text data.

In an embodiment, when the processor 1001 performs the above-mentioned conversion of each service voice data in the at least one service voice data into service text data, the following steps are specifically performed:

selecting one service voice data from the at least one service voice data as target voice data, and acquiring voice spectrum characteristics in the target voice data;

performing voiceprint recognition on the target voice data based on the voice frequency spectrum characteristics and a first recognition model to obtain a voiceprint recognition result corresponding to the first recognition model; the voiceprint recognition result comprises the matching degree between the voice spectrum characteristic and a plurality of attribute spectrum characteristics in the first recognition model;

according to the voiceprint recognition result, using the label information associated with the attribute spectrum feature with the highest matching degree with the voice spectrum feature as a first role recognized from the target voice data;

determining the rest roles except the first role in the target voice data as second roles, and converting the target voice data carrying the first role and the second role into service text data based on a control frequency parameter and a second identification model;

and when each service voice data in the at least one service voice data is taken as the target voice data, obtaining service text data corresponding to each service voice data.

In an embodiment, when the processor 1001 obtains associated text information associated with the target text information from the target service text data and performs the abnormality detection on the associated text information according to an associated quality inspection point in the target quality inspection rule, the following steps are specifically performed:

taking the target quality inspection point corresponding to the first identifier as a starting inspection point, and acquiring a detection range threshold corresponding to the target quality inspection rule;

determining text information between the starting checkpoint and the detection range threshold as associated text information associated with the target text information;

and acquiring associated quality inspection points in the target quality inspection rule, wherein the associated quality inspection points have an association relation with the target quality inspection points, and performing anomaly detection on the associated text information according to the associated quality inspection points.

In one embodiment, the processor 1001 may further perform:

adding the target abnormal text to an abnormal text data set, and sending all target abnormal texts in the abnormal text data set to a quality inspection terminal;

and acquiring abnormal quality inspection points determined by the quality inspection terminal based on all target abnormal texts, and updating the target quality inspection rule based on the abnormal quality inspection points.

In one embodiment, the processor 1001 may further perform:

adding the target abnormal text to an abnormal text data set, acquiring historical abnormal text corresponding to the abnormal text data set, and obtaining a training sample set based on the historical abnormal text;

training a quality inspection model corresponding to the target quality inspection rule based on the training sample set, and updating the quality inspection model; and the abnormal quality inspection points carried in the updated quality inspection model are used for updating the target quality inspection rule.

It should be understood that the data processing apparatus 1000 described in the embodiment of the present invention may perform the description of the data processing method in the embodiment corresponding to any one of fig. 2 to fig. 7, and may also perform the description of the data processing apparatus 1 in the embodiment corresponding to fig. 8, which is not repeated herein. In addition, the beneficial effects of the same method are not described in detail.

Furthermore, it is to be noted here that: an embodiment of the present invention further provides a computer storage medium, where the computer storage medium stores the aforementioned computer program executed by the data processing apparatus 1, and the computer program includes program instructions, and when the processor executes the program instructions, the description of the data processing method in any one of the embodiments of fig. 2 to fig. 7 can be executed, so that details are not repeated here. In addition, the beneficial effects of the same method are not described in detail. For technical details not disclosed in the embodiment of the computer storage medium related to the present invention, refer to the description of the embodiment of the method of the present invention.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by a computer program, which can be stored in a computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), or the like.

The above disclosure is only for the purpose of illustrating the preferred embodiments of the present invention, and it is therefore to be understood that the invention is not limited by the scope of the appended claims.

Claims

1. A data processing method, comprising:

acquiring target service text data, acquiring target text information from the target service text data, and performing anomaly detection on the target text information according to a target quality inspection rule in a target quality inspection rule;

if the associated text information has associated quality inspection points meeting the target quality inspection rule, setting a second identifier corresponding to the associated quality inspection points in the associated text information, and determining target service text data carrying the first identifier and the second identifier as a target abnormal text;

wherein, the obtaining of the associated text information associated with the target text information from the target service text data and the abnormal detection of the associated text information according to the associated quality inspection point in the target quality inspection rule include:

determining text information between the starting checkpoint and the detection range threshold as associated text information associated with the target text information; the associated text information is text information between the first detection range thresholds before the starting check point when the detection range thresholds include the first detection range threshold; or the associated text information is text information between the second detection range thresholds after the starting checkpoint when the detection range thresholds include the second detection range threshold; or the associated text information is text information between the first detection range threshold before the start checkpoint and text information between the second detection range threshold after the start checkpoint when the detection range threshold includes the first detection range threshold and the second detection range threshold;

2. The method of claim 1, wherein before obtaining the target service text data, further comprising:

the method comprises the steps of obtaining at least one service voice data, respectively converting each service voice data in the at least one service voice data into service text data, and selecting one service text data from the converted service text data as target service text data.

3. The method of claim 2, further comprising:

counting the target quantity of all the service text data obtained by conversion;

4. The method of claim 2, wherein the converting each service voice data of the at least one service voice data into service text data comprises:

5. The method of claim 4, wherein the obtaining target text information from the target service text data and performing anomaly detection on the target text information according to a target quality inspection point in a target quality inspection rule comprises:

acquiring first text data corresponding to the first role and second text data corresponding to the second role from the target service text data;

and acquiring a target quality inspection rule from the plurality of quality inspection rules, taking the first text data and/or the second text data as target text information, selecting a quality inspection point with the highest priority from the plurality of quality inspection points contained in the target quality inspection rule as a target quality inspection point, and carrying out anomaly detection on the target text information based on the target quality inspection point.

6. The method of claim 1, further comprising:

7. The method of claim 1, further comprising:

8. A data processing apparatus, comprising:

a second setting module, configured to set a second identifier corresponding to the associated quality inspection point in the associated text information if the associated quality inspection point in the target quality inspection rule is met in the associated text information, and determine target service text data carrying the first identifier and the second identifier as a target abnormal text;

wherein, the associated information acquisition module comprises:

the check point determining unit is used for taking the target quality check point corresponding to the first identifier as a starting check point and acquiring a detection range threshold corresponding to the target quality check rule;

a first determination unit configured to determine text information between the start checkpoint and the detection range threshold as associated text information associated with the target text information; the associated text information is text information between first detection range thresholds before the starting checkpoint when the detection range thresholds include the first detection range threshold; or the associated text information is text information between the second detection range thresholds after the starting checkpoint when the detection range thresholds include the second detection range threshold; or the associated text information is text information between the first detection range threshold before the start checkpoint and the second detection range threshold after the start checkpoint when the detection range threshold includes the first detection range threshold and the second detection range threshold;

and the abnormality detection unit is used for acquiring the associated quality inspection points which have an associated relation with the target quality inspection points in the target quality inspection rule and carrying out abnormality detection on the associated text information according to the associated quality inspection points.

9. A data processing apparatus, comprising: a processor and a memory;

the processor is coupled to a memory, wherein the memory is configured to store program code and the processor is configured to invoke the program code to perform the method of any of claims 1-7.