CN115358774A - Data processing method and system for intelligent customer service robot - Google Patents

Data processing method and system for intelligent customer service robot Download PDF

Info

Publication number
CN115358774A
CN115358774A CN202210974456.2A CN202210974456A CN115358774A CN 115358774 A CN115358774 A CN 115358774A CN 202210974456 A CN202210974456 A CN 202210974456A CN 115358774 A CN115358774 A CN 115358774A
Authority
CN
China
Prior art keywords
robot
customer service
intelligent customer
index
service robot
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210974456.2A
Other languages
Chinese (zh)
Inventor
于霄洋
郑蓉蓉
王晨辉
薛文婷
武志栋
曾诣佳
朱京
冯显时
李雅西
李枫
刘娇丽
闫瑜
王蕊
邹晓颖
许大卫
罗大勇
杜加文
张伟
蒋一凡
彭苒
宫晓峰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Henan Jiuyu Tenglong Information Engineering Co ltd
State Grid Corp of China SGCC
State Grid Information and Telecommunication Co Ltd
Original Assignee
Henan Jiuyu Tenglong Information Engineering Co ltd
State Grid Corp of China SGCC
State Grid Information and Telecommunication Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Henan Jiuyu Tenglong Information Engineering Co ltd, State Grid Corp of China SGCC, State Grid Information and Telecommunication Co Ltd filed Critical Henan Jiuyu Tenglong Information Engineering Co ltd
Priority to CN202210974456.2A priority Critical patent/CN115358774A/en
Publication of CN115358774A publication Critical patent/CN115358774A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0281—Customer communication at a business location, e.g. providing product or service information, consulting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/23—Updating
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/332—Query formulation
    • G06F16/3329—Natural language query formulation or dialogue systems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/3331—Query processing
    • G06F16/3332—Query translation
    • G06F16/3338—Query expansion
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/279—Recognition of textual entities
    • G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/30—Semantic analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00—Speech recognition
    • G10L15/01—Assessment or evaluation of speech recognition systems
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00—Speech recognition
    • G10L15/06—Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
    • G10L15/063—Training
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00—Speech recognition
    • G10L15/08—Speech classification or search
    • G10L15/18—Speech classification or search using natural language modelling
    • G10L15/1822—Parsing for meaning understanding
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00—Speech recognition
    • G10L15/22—Procedures used during a speech recognition process, e.g. man-machine dialogue
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
    • G10L25/48—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use
    • G10L25/51—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use for comparison or discrimination
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00—Speech recognition
    • G10L15/06—Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
    • G10L15/063—Training
    • G10L2015/0635—Training updating or merging of old and new templates; Mean values; Weighting

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Computational Linguistics (AREA)
  • General Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Human Resources & Organizations (AREA)
  • Artificial Intelligence (AREA)
  • Strategic Management (AREA)
  • General Engineering & Computer Science (AREA)
  • Human Computer Interaction (AREA)
  • Acoustics & Sound (AREA)
  • Economics (AREA)
  • Development Economics (AREA)
  • Multimedia (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • General Business, Economics & Management (AREA)
  • Mathematical Physics (AREA)
  • Game Theory and Decision Science (AREA)
  • Databases & Information Systems (AREA)
  • Marketing (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • Educational Administration (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Signal Processing (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses a data processing method and a system of AN intelligent customer service robot, which comprises the steps of calculating the speech recognition single word accuracy rate A, the semantic recognition accuracy rate N, the whole service accuracy rate P, the problem solving rate S, the independent reception duty ratio P of the robot and the basic index values of other robots of the intelligent customer service robot, and establishing a service quality general index AN-PSR of the intelligent customer service robot; setting a self-defined index by a user according to the basic data; evaluating the performance of the intelligent customer service robot according to the set AN-PSR index threshold or the user-defined index threshold and the intelligent customer service robot index value to generate AN optimization strategy; the system comprises a robot key index acquisition module, a robot optimization suggestion module and a robot self-defined index acquisition module. The method sets the evaluation index of the overall service quality of intelligent customer service and provides the optimization index according to the index. And the machine is provided for automatically updating the database and the knowledge base, and the optimization iteration of the intelligent customer service robot is completed.

Description

Data processing method and system for intelligent customer service robot
Technical Field
The invention relates to the technical field of evaluation methods, in particular to a data processing method and system of an intelligent customer service robot.
Background
The customer service can be divided into manual customer service and electronic customer service, wherein the manual customer service can be divided into three types of character customer service, video customer service and voice customer service. The character customer service is customer service mainly in a typing chat mode; the video customer service means that customer service is mainly carried out in a voice video form; voice service refers to service that is performed primarily in the form of a mobile phone. There are many customer service methods and systems in the prior art, for example, patent document CN109615407A, which discloses an online intelligent customer service method for electric power based on artificial intelligence, comprising: receiving the consultation information of the user, and carrying out language analysis processing on the consultation information of the user by adopting artificial intelligence; extracting characteristic information of the processed consultation information; matching the characteristic information with the answer information in the knowledge database, and pushing the matched answer information to the user if the matching result exists; if the matching result does not exist, the manual processing is converted, and meanwhile, the answer information of the manual processing corresponds to the current characteristic information and is updated into a knowledge database; the system can accurately identify the problem intention of the user aiming at the consultation information of the user, make accurate solution and push, and can change the system into manual solution when the customer service system cannot automatically solve, and can timely update the information in the knowledge database after the manual solution, thereby providing guarantee for the efficiency and accuracy of subsequent customer service.
The existing intelligent customer service robot has low data processing efficiency, so the application aims to provide a data processing method and a data processing system of the intelligent customer service robot, which can overcome the defects.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a data processing method and a data processing system of an intelligent customer service robot.
The purpose of the invention is realized by the following technical scheme: an intelligent customer service robot data processing method comprises the following steps: s100, after asking a question to the smart customer service robot, the system carries out basic data statistics and index detection on the robot according to self-timing based on the answer condition of the smart customer service robot, and calculates the robot voice recognition single word accuracy rate A, the semantic recognition accuracy rate N, the whole service accuracy rate P, the problem solution rate S and the robot independent reception occupation ratio P, and waits for self-defined index information; s200, establishing a service quality general index AN-PSR of the intelligent customer service robot or AN index value of the intelligent customer service robot calculated according to a self-defined index formula, setting a threshold value, evaluating the performance of the intelligent customer service robot, and generating AN optimization strategy when the evaluation level of the intelligent customer service robot is lower than the set level; s300, when the index of the intelligent customer service robot is lower than a set threshold value, the optimization strategy can optimize the service quality of the robot and the overall index AN-PSR or the user-defined index of the service quality, so that the service capacity of the intelligent customer service robot is improved.
Preferably, the total quality of service indicator AN-PSR can be represented by AN-PSR = w (x × a + (1-x) × N) + (1-w) × (y × C + z × S + (1-y + z) × R), where x represents a ratio of speech recognition individual word accuracy to semantic understanding accuracy in the technical indicator according to actual traffic analysis; y represents the proportion of the accuracy in the service indexes to the total service indexes according to the actual service analysis; z represents the ratio of the problem resolution in the service index to the overall service index according to the actual service analysis.
Preferably, the optimization strategy includes: adding manual intervention into the original ASR speech recognition model, re-screening the intelligent customer service robot by using the accuracy of manual information, and repeatedly training the data of the dry prognosis so as to update the ASR speech recognition model, so that the accuracy A of the speech recognition single character can be improved.
Preferably, the optimization strategy further comprises: and creating a customer service keyword database, and adding the customer service keyword database into a semantic understanding analysis model, so that when a customer asks about a keyword in the customer service keyword database, the keyword is not segmented by a word segmentation model, and the semantic recognition accuracy rate N is improved.
Preferably, the optimization strategy further comprises: accumulating knowledge according to historical consultation data, dividing the knowledge into high, medium and low frequencies, and preferentially covering high-frequency consultation problems; when a new business policy is involved, a knowledge point is actively added, and the knowledge point is regularly sorted to check whether the intercommunicated business inquiry methods exist in different business lines; the intelligent customer service robot question-returning function is set, when a customer asks knowledge points of several service intercommunicating, the intelligent customer service robot asks the knowledge points, so that the customer can accurately find question classification to ask.
Preferably, the overall service accuracy P = response coverage × coverage accuracy, where response coverage = number of responses/total number of triggers accepted by the robot, and the accuracy in coverage = number of accurate responses/number of total responses of the robot.
Preferably, the step of obtaining the index value of the intelligent customer service robot comprises the following steps: the intelligent customer service robot receives the questions sent by the user, translates the voice information into text information and semantically understands the text information, replies corresponding answers to the user according to the set answering strategies, conducts reciprocating operation until the whole round of conversation is completed, and obtains and calculates the specified indexes in the process.
The application also provides an intelligence customer service robot data processing system, its characterized in that includes: the robot key index acquisition module is used for inputting a plurality of questions to be answered to the intelligent customer service robot, carrying out index detection based on the answering conditions of the intelligent customer service robot, and calculating the robot voice recognition single character accuracy rate A, the semantic recognition accuracy rate N, the overall service accuracy rate P, the question resolution rate S, the robot independent reception duty ratio P and the robot basic index value; establishing a service quality general index AN-PSR of the intelligent customer service robot or self-defining AN index according to a use habit, evaluating the performance of the intelligent customer service robot by the index, and generating AN optimization strategy when the evaluation level of the intelligent customer service robot is lower than a set level; when the intelligent customer service robot index is lower than a set threshold value, the optimization strategy can optimize the service quality of the robot and the service quality general index AN-PSR or the user-defined index.
Preferably, the method further comprises the following steps: the robot database module is used for storing and calling related information such as a corpus of the daily robot, data of user question information, a robot training model and the like; the user information understanding module is used for translating the problems of the user into a character form by utilizing a voice recognition technology; the semantic understanding module is used for understanding the problems of the user, identifying the intentions of the problems of the user and outputting the content of an identification result; and the data information display module is used for displaying the key data and the administrator personalized data.
Preferably, when the evaluation level of the intelligent customer service robot is lower than a set level, an optimization strategy is generated, and the optimization strategy can optimize the robot database module and the semantic understanding module.
The application also provides a computer-readable storage medium, which stores computer-executable instructions, and when the instructions are executed by a processor, the instructions implement any one of the steps of the data processing method of the intelligent customer service robot.
The application also provides a computer program, wherein when the computer program is executed in a computer, the computer is caused to execute the steps of the intelligent customer service robot data processing method. The invention has the following advantages:
the method is based on indexes, an intelligent customer service quality optimization system is established, overall service quality evaluation indexes of intelligent customer service are set, and the method for optimizing the indexes is provided according to the indexes. And the machine is provided to automatically update the database and the knowledge base, and the optimization iteration of the intelligent customer service robot is completed.
(1) The method and the system establish and sum the indexes of the prior art, and perform targeted performance improvement according to the indexes to form an intelligent customer service quality evaluation and technical service performance improvement system.
(2) The method and the system have the advantages that service indexes of various performances of the robot are created on the background of improving the service quality of the intelligent customer service robot; the specific promotion scheme of each index is perfected; a full-automatic flow for promoting the intelligent customer service robot is formed; a complete set of intelligent customer service robot quality improvement system is formed.
Drawings
FIG. 1 is a schematic flow chart illustrating a data processing method for an intelligent customer service robot according to the present invention;
FIG. 2 is a schematic diagram of a modular structure of the intelligent customer service robot data processing system according to the present invention.
In the figure, the system comprises a robot key index acquisition module 1, a robot custom index acquisition module 2, a robot optimization suggestion module 3, a user information understanding module 4, a data information display module 5, a robot database module 6 and a semantic understanding module 7.
Detailed Description
The invention will be further described with reference to the accompanying drawings, without limiting the scope of the invention to the following:
as shown in fig. 1 and fig. 2, the present application provides an intelligent customer service robot data processing system, which includes a robot key index acquisition module 1, a robot custom index acquisition module 2, a robot optimization suggestion module 3, a user information understanding module 4, a data information display module 5, a robot database module 6, and a semantic understanding module 7. When the intelligent customer service robot operates normally, the robot key index acquisition module 1 can read indexes such as robot voice recognition single character accuracy, semantic recognition accuracy, overall service accuracy, problem solving rate, robot independent reception duty ratio and the like, and dynamically updates index data in real time. Specifically, the speech recognition individual word accuracy =1- (number of inserted error words + number of deleted error words + number of replaced error words)/number of total words of the labeling result × 100% (· number of inserted error words: understandable to multi-transcribed (number of words that are not heard by man but transcribed by machine); number of deleted error words: understandable to few transcribed (number of words that are heard by man but not transcribed by machine); number of replaced error words: understandable to transcribed by man (number of words that are transcribed by man but transcribed by machine)). Keyword accuracy =1- (number of inserted error keywords + number of deleted error keywords + number of replaced error keywords)/total number of keywords of tagging result × 100%. (number of inserted error keywords: number of understandable multiple transcription (number of keywords that were not heard manually but transcribed by machine); number of deleted error keywords: number of understandable fewer transcription (number of keywords that were heard manually but not transcribed by machine); number of replacement error keywords: number of understandable transcription errors (number of keywords that were heard manually but transcribed by machine)). Single sentence accuracy =1 — correct message volume/total message volume is transcribed. Knowledge hit ratio =1 — ratio of bottom-to-bottom answer (bottom-to-bottom answer means that the robot does not understand the meaning of a human and enters a set bottom-to-bottom sentence). Intended accuracy: and (3) carrying out batch test on real data of the machine by using a real data batch test, comparing the intention recognized by the robot with the intention recognized by the worker marking person with a worker marking graph, comparing the manual with the machine classification graph, and judging that the intention is recognized accurately if the manual and machine classifications are consistent, otherwise, judging that the intention is recognized accurately, and otherwise, judging that the intention is recognized wrongly. Response accuracy: by marking the real log on the line, the robot responds the message that the robot intends to identify correctly, and the answer of the robot can solve the problem proposed by the user. Overall traffic accuracy = response coverage accuracy. Response coverage = number of responses/number of robot take-over total triggers. Accuracy in coverage = number of exact responses/number of total responses of the robot. The problem resolution ratio = the amount of questions of the customer to be completely answered by the intelligent customer service robot/the amount of questions of all the customers to be served by the intelligent customer service robot. The problem resolution is a comprehensive index and a high-order index, and two calculation modes are generally adopted in the industry: (1) judging whether the problem of the user is solved or not by researching a questionnaire form after the call; (2) by judging whether the user repeatedly consults with the incoming line within 24 hours; if the user repeatedly enters the line, the user may not be solved before the problem is solved and the line is consulted again. The robot is in independent reception ratio: in the one-pass conversation, the robot receives the conversation independently completely, and the ratio conversion manual rate without conversion manual processing = conversion manual conversation volume/all consultation conversation volume.
Preferably, the robot-defined index obtaining module 2 uses basic data such as receptions and answers generated by the robot to freely create a personalized index that the user wants to create, generates data of a new index on the system according to mathematical operations among the basic data, and calculates historical data according to the generated index. For example, the system gives basic data of the robot, such as work order quantity, call return turns, user evaluation rate, robot return times and the like. The manager can carry out self-defined indexes of basic mathematical operations such as addition, subtraction, multiplication, division, power and the like on the basic data. Also for example, AN overall quality of service indicator AN-PSR of the intelligent customer service robot can be established. And the intelligent customer service robot system is used for evaluating the performance of the intelligent customer service robot according to a set threshold and the index value of the intelligent customer service robot and generating an optimization strategy when the evaluation level of the intelligent customer service robot is lower than the set level. For example, the administrator sets a threshold value for each index, and when the index of the robot is lower than the threshold value, the robot optimization suggestion module 3 promotes the operation suggestion of the robot to the administrator according to the information represented by the index. The robot database module 6 is used for storing and calling the corpus of the daily robot, the data of the user question information, the robot training model and other related information. The user information understanding module 4 can translate the user question into a character form by using a voice recognition technology, understand the user question by using the semantic understanding module 7, recognize the intention of the user question and output the recognition result content. The data information display module 5 can display the key data and the administrator personalized data, and is provided with a data information query function.
Preferably, the intelligent customer service robot data processing system of the present application is configured to evaluate and optimize the performance of the intelligent customer service robot according to the following steps:
s100, after asking questions of the intelligent customer service robot, the system carries out basic data statistics and index detection on the robot according to self-timing based on the answer condition of the intelligent customer service robot, calculates the speech recognition single word accuracy rate, the semantic recognition accuracy rate, the overall service accuracy rate, the problem solving rate and the robot independent reception duty ratio, and waits for self-defined index information.
Specifically, the intelligent customer service robot receives questions sent by a user, the intelligent customer service robot translates voice information into text information and carries out semantic understanding on the text information, the intelligent customer service robot replies corresponding answers to the user according to the set answering strategies, and the intelligent customer service robot operates repeatedly until the whole round of conversation is completed. Aiming at the process, the robot key index acquisition module 1 can acquire and calculate the specified index, so that the index value of the intelligent customer service robot is acquired.
S200, calculating to obtain an index value of the intelligent customer service robot according to a self-defined index formula, setting a threshold value, evaluating the performance of the intelligent customer service robot, and generating an optimization strategy when the evaluation level of the intelligent customer service robot is lower than the set level.
Specifically, the robot database module and the semantic understanding module are optimized according to the solving strategies corresponding to all indexes by judging the current level of the robot. The set threshold comprises a normal operation threshold and a threshold needing to be optimized. Wherein the normal operation threshold value represents the performance of the current intelligent customer service robot without optimization, and the intelligent customer service function can be normally completed. The threshold value to be optimized represents that the level of the current intelligent customer service robot is poor, the current user conversation function cannot be completed, and the optimization needs to be carried out immediately. And judging the relation between the index of the intelligent customer service robot and the threshold at regular time, and when the index of the intelligent customer service robot is higher than the normal operation threshold, representing that the intelligent customer service robot operates normally without changing the content of the intelligent customer service robot. When the index value of the intelligent customer service robot is lower than the normal operation value but higher than the threshold value to be optimized, the intelligent customer service robot represents that partial contents of the intelligent customer service robot have problems, but can still normally operate at present. When the intelligent customer service robot is lower than the threshold value to be optimized, the intelligent customer service robot represents that the positive optimization needs to be carried out on the index with the poor numerical value.
Preferably, the optimization of the robot database module and the speech understanding module includes at least the following aspects:
(1) And adding manual intervention into the original ASR speech recognition model, performing data screening on the intelligent customer service robot again by using the accuracy of manual information, and performing repeated training on the data after the intervention so as to update the ASR speech recognition model. Specifically, when the speech recognition accuracy is lower than the set threshold, the system needs to send out a warning to remind the manager that the part needs to be optimized. A manager needs to check the voice recognition content at the translation position of the robot, if an error part exists, the manual correction is needed, the deep learning training is carried out on the change result again to obtain a new voice recognition model, and the accuracy rate of the voice recognition of single characters can be improved through the method.
(2) Creating a customer service keyword database, adding the customer service keyword database into a semantic understanding analysis model, and performing word segmentation processing on sentences by the semantic recognition model according to needs, wherein the word segmentation refers to segmenting a customer service natural language text into word sequences. When a client asks for a keyword in the customer service keyword database, the segmentation model does not segment the keyword any more so as to improve the semantic recognition accuracy. Specifically, the intelligent customer service robot obtains user question contents, performs word segmentation on each piece of question contents, and performs ranking display of the number of question times on a large number of corpus word segmentation results at regular time. The manager sets the adding quantity and frequency of the word stock in advance, and the system automatically adds words with more times of occurrence of the keywords to the national network communication customer service keyword stock at regular time according to the set numerical value and frequency of the manager. When the semantic understanding rate is lower than the set threshold value to be optimized, the system automatically learns the national network communication customer service word stock after the keywords are added again, generates a new semantic understanding analysis model, and circularly detects indexes of the model until the indexes reach the normal threshold value.
Preferably, the semantic understanding analysis model is trained according to the following steps:
s1: the data preparation method specifically comprises the following steps: 1) And putting the data of each label into different folders, merging one folder into one class, and counting the number of each label. 2) And sample equalization, namely equalizing the belt training samples to each folder. 3) Segmenting the sample set, for example: 90% are used for training, 10% are left for testing, and the proportion is determined by a manager. For the weak classes to be resampled, the test set does not need to be resampled. In training, the sample balance is ensured, the characteristics of the weakness class are learned, and the real data set distribution is reflected in the testing process.
S2: the data processing method specifically comprises the following steps: 1) After the training data is generated, word lists corresponding to all categories in the training data need to be obtained. And performing word segmentation on the linguistic data asked by the user in each folder by using a jieba word segmentation tool, removing stop words, and finally counting the occurrence frequency of similar words in the folder to form an embedded matrix M. 2) The text is converted to id using a vocabulary. And limit the maximum length of the sentence, and the length is insufficient to complement 0. The output result content is as follows: the first number in each line is the category, and the remaining numbers are the sentence content id.
S3: and (3) model training, namely building a convolution layer, a pooling layer, a full-link layer and a softmax layer.
1) Convolutional layer
First, a sentence is input, word segmentation is carried out, and s words are assumed to be obtained. Then, according to the obtained embedding matrix M, word vectors can be integrated. Assume that the word vector has d dimensions. Then for the input sentence a matrix a e Rs x d of s rows and d columns is obtained.
Considering matrix a as an image, a convolutional neural network is used to extract the features of a. Here, since the relevance of adjacent words in a sentence is always high, the width of the convolution kernel is the dimension d of the word vector, and the height is a settable hyper-parameter. If the convolution kernel is a matrix w with width d and height h, then w has d x h parameters to be updated. For a sentence there is a matrix A ∈ Rs × d. a [ i: j ] denotes the ith row and jth column of A. The convolution operation can be expressed as oi = w · a [ i: i + h-1], i =1,2. Then, the bias b is superposed, and the activation function f is used for activation, so that the required characteristics are obtained, and the formula is as follows: ci = f (oi + b). For a convolution kernel, the characteristic c belongs to Rs-h +1, and the total of s-h +1 characteristics can be obtained. Specific experiments to obtain a richer semantic representation of the features, a plurality of convolution kernels of different heights may be used.
2) A pooling layer: for the final feature extraction, the convolution kernels need to be made dimensionally the same. The maximum value in the feature map needs to be extracted using a 1-max Pooling pooling function for each feature map.
3) Full connection layer: for all convolution kernels: 1-max pooling functions, which are then cascaded, the final number of prediction classes can be obtained: num _ class.
4) softmax layer: and (5) splicing the max-posing results, and sending the result into softmax to obtain the probability that each category, such as label, is 1 and the probability that label is-1. And calculating a loss function according to the predicted label and the actual label, and calculating gradients of parameters needing to be updated in the four functions, namely the softmax function, the max-posing function, the activation function and the convolution kernel function, so as to sequentially update the parameters in the four functions and finish a round of training. And repeating the steps until the loss function training is minimum, and finishing the training.
5) And predicting by using the test set, and evaluating the accuracy on the test set or the verification set. And repeating the steps according to the reaction result of the test set.
(3) Accumulating knowledge according to historical consultation data, dividing the knowledge into high, medium and low frequencies, and preferentially covering the high-frequency consultation problem; when a new business policy is involved, a knowledge point is actively added, and the knowledge point is regularly sorted to check whether the intercommunicated business inquiry methods exist in different business lines; the intelligent customer service robot question-returning function is set, when a customer asks knowledge points of several service intercommunicating, the intelligent customer service robot asks the knowledge points, so that the customer can accurately find question classification to ask. Specifically, the system classifies the user question, the deep learning module automatically understands the user question and ranks the questions to form a ranking of the number of popular questions. When the accuracy of the overall service is improved to be lower than the set threshold value to be optimized, the system adds the hot problem into the knowledge base, carries out index detection on the updated knowledge base, and circulates the steps until the index reaches the normal threshold value. The system will automatically identify whether there are similar question questions in the user's question categories. When a plurality of scenes are identified to have similar questioning methods, a question-back system can be automatically formed, when a user asks the question, the intelligent customer service robot asks the user to find out which scene the user asks, and after the user selects, the user answers the question in detail according to the scenes.
(4) The similar question expansion of each question is particularly important, a plurality of questions are asked by a client for one question, and when the client does not ask the standard question set by the robot, the robot cannot judge the question of the user, so that the problem solving rate of the user is correspondingly reduced. And when the user problem solving rate is lower than the set threshold value to be optimized, the system expands according to the existing user problems in the knowledge base. And automatically performing similarity question expansion on the user questions through a deep learning algorithm, and adding the similarity questions into a knowledge base. The system detects the problem resolution value of the current user in real time, and stops the expansion of the similarity problem when the value reaches a normal threshold value.
(5) The main factor influencing the independent reception duty ratio of the robot is that the reserve capacity of the knowledge base of the robot is insufficient, and when the robot does not have enough answering conversations, the answering efficiency is greatly reduced. Scene classification: the manager needs to sort the service classes, clearly has a plurality of service lines or service scenes, and sorts the service lines or the service scenes from coarse to fine. Removing confusion: in the operation process, whether the combed content is repeated or not needs to be paid attention, if multi-person combing or multi-person operation is involved, a newly added knowledge point synchronization mechanism needs to be established, and the problem that the robot answers questions illegally due to repeated confusion is avoided. Knowledge label: in the carding process, knowledge points need to be marked, which knowledge points can be responded by a robot, which parts (difficulty and emotion) need to be responded by a human, and the identification is needed in the early stage of carding. The system can be used for synchronizing the offline knowledge base, and the content of the offline knowledge base is corresponding to the content of the near-intelligent customer service robot system. When a new scenario problem is encountered, the system automatically recognizes and creates a new scenario in the system. And carrying out knowledge point confusion detection with other scene problems. When knowledge confusion occurs in the same scene, the system can warn a manager to perform detection, and the optimal solution of the user question is reserved. When knowledge confusion problems occur in different scenes, the system can automatically form a question language in the robot question answering. When the user asks the question, the robot lets the user choose which scenario to ask. When the independent reception duty ratio value of the robot is lower than the threshold value to be optimized, the system can immediately synchronize the off-line knowledge base and optimize the structure and the content of the system knowledge base. The system detects the independent receptivity occupation ratio value of the robot in real time, and when the value is higher than a normal threshold value, the synchronization of the knowledge base is stopped.
(6) The manager sets the solution corresponding to the index, and the solution comprises newly joining the semantic understanding analysis model, retraining the voice recognition model, optimizing the knowledge base and the like. And when the self-defined index is lower than the threshold value to be optimized, the system automatically optimizes according to the strategy. If the optimization needs policy manual intervention, the system sends an alarm to remind a manager to perform manual optimization.
S300, when the indexes of the intelligent customer service robot are lower than a set threshold value, the optimization strategy can optimize the service quality overall index AN-PSR or the self-defined index, so that the service capacity of the intelligent customer service robot is improved, wherein the service quality overall index AN-PSR can be represented by AN-PSR = w (x A + (1-x) N) + (1-w) x (y C + z S + (1-y + z) R), and the technical index is abbreviated as voice recognition (ASR) to A; semantic understanding (NLU) is N; the overall service accuracy (precision) is abbreviated as P; the problem solving rate (Solve) is abbreviated as S; the robot-created-person independent reception (Peception) ratio is abbreviated as R. w represents a ratio numerical value of the technical capacity in the real service to the whole capacity of the technology and the service; x represents the ratio of the speech recognition single character accuracy rate and the semantic understanding accuracy rate in the technical indexes according to the actual business analysis; y represents the proportion of the accuracy in the service indexes to the total service indexes according to the actual service analysis; and z represents the proportion of the problem solution rate in the service index to the whole service index according to the actual service analysis. In semantic understanding (NLU) metrics: the intention accuracy rate is the probability that the intention identified by the robot is always consistent with the manual marking intention; the response accuracy rate refers to the probability of accurate response in the message responded by the robot; knowledge hit rate = (1-paphiopedial response ratio), which means the ratio of robot understanding human intent first, making accurate response second, and subtracting number of entries without response with 1 finally. Therefore, the level of the semantic recognition knowledge hit rate is higher than the intention accuracy rate and the response accuracy rate, so the knowledge hit rate is selected as the technical index of semantic understanding. Speech Recognition (ASR) metrics: the individual character accuracy rate refers to the ratio of correctly translated words in a sentence; the hot word accuracy rate refers to the proportion of the number of hot words with correct translation in a sentence; the single sentence accuracy rate refers to the ratio of the number of correctly translated sentences in the robot answer to the whole answer. The concept of single sentence accuracy includes single word accuracy and hot word accuracy. Therefore, the sentence accuracy (which may be abbreviated as ASR accuracy) is selected as the technical index of speech recognition.
Embodiments of the present application also provide a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement any of the steps of the method for processing data of an intelligent customer service robot.
The embodiment of the present application further provides a computer program, wherein when the computer program is executed in a computer, the computer is caused to execute the steps of the intelligent customer service robot data processing method.
Finally, it should be noted that: although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that changes may be made in the embodiments and/or equivalents thereof without departing from the spirit and scope of the invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (10)

1. An intelligent customer service robot data processing method is characterized by comprising the following steps:
s100, after questions are asked for the intelligent customer service robot, the system performs basic data statistics and index detection on the robot according to self-timing based on the answer condition of the intelligent customer service robot, and calculates self-defined index information related to the robot voice recognition single character accuracy A, the semantic recognition accuracy N, the overall service accuracy P, the problem solution rate S and the robot independent reception duty ratio P;
s200, establishing a service quality general index AN-PSR of the intelligent customer service robot or AN index value of the intelligent customer service robot calculated according to a self-defined index formula, setting a threshold value, evaluating the performance of the intelligent customer service robot, and generating AN optimization strategy when the evaluation level of the intelligent customer service robot is lower than the set level;
s300, when the index of the intelligent customer service robot is lower than a set threshold value, the optimization strategy can optimize the service quality of the robot and the overall index AN-PSR or the user-defined index of the service quality, so that the service capacity of the intelligent customer service robot is improved.
2. The intelligent customer service machine data processing method according to claim 1, wherein the overall quality of service indicator AN-PSR can be represented by AN-PSR = w (x a + (1-x) N) + (1-w) C + z S + (1-y + z) R), where x represents a ratio of speech recognition individual word accuracy and semantic understanding accuracy in the technical indicator according to actual traffic analysis; y represents the proportion of the accuracy in the service index to the total service index according to the actual service analysis; z represents the ratio of the problem resolution in the service index to the overall service index according to the actual service analysis.
3. The intelligent customer service machine data processing method of claim 2 wherein the optimization strategy includes: and adding manual intervention into the original ASR speech recognition model, carrying out data screening on the intelligent customer service robot again by utilizing the accuracy of manual information, and repeatedly training the data of the dry-out state so as to update the ASR speech recognition model, so that the accuracy A of the speech recognition single character can be improved.
4. The intelligent customer service machine data processing method of claim 2 wherein the optimization strategy further comprises: and creating a customer service keyword database, and adding the customer service keyword database into a semantic understanding analysis model, so that when a customer asks for a keyword in the customer service keyword database, a word segmentation model does not divide the keyword, and the semantic recognition accuracy rate N is improved.
5. The intelligent customer service machine data processing method of claim 2 wherein the optimization strategy further comprises: accumulating knowledge according to historical consultation data, dividing the knowledge into high, medium and low frequencies, and preferentially covering the high-frequency consultation problem; when a new policy of a new service is involved, actively adding a knowledge point, and regularly sorting the knowledge point to check whether an intercommunicated service question exists in different service lines; the intelligent customer service robot question-returning function is set, when a customer asks knowledge points of several service intercommunicating, the intelligent customer service robot asks the knowledge points reversely, so that the customer can accurately find question classification to ask.
6. The method as claimed in claim 1, wherein the overall business accuracy P = answer coverage rate × coverage accuracy, wherein answer coverage rate = number of answers/number of overall triggers accepted by the robot, and wherein accuracy in coverage = number of answers accurate/number of overall responses of the robot.
7. The intelligent customer service robot data processing method according to claim 1, wherein obtaining the index value of the intelligent customer service robot comprises the steps of: the intelligent customer service robot receives the questions sent by the user, translates the voice information into text information and carries out semantic understanding on the text information, replies corresponding answers to the user according to the set answering strategies, operates repeatedly until the whole round of conversation is completed, and obtains and calculates the specified indexes in the process.
8. An intelligent customer service robot data processing system, comprising:
the robot key index acquisition module (1) is used for inputting a plurality of questions to be answered to the intelligent customer service robot, carrying out index detection based on the answering situation of the intelligent customer service robot, and calculating the speech recognition single word accuracy rate A, the semantic recognition accuracy rate N, the overall service accuracy rate P, the question resolution rate S and the robot independent reception duty ratio P so as to acquire an index value of the intelligent customer service robot;
the robot self-defined index acquisition module (2) freely creates the personalized index which the user wants to create by using the basic data such as receptions, answers and the like generated by the robot, generates data of a new index on the system according to mathematical operation among the basic data, and calculates historical data according to the generated index. For example, the system gives basic data of the robot, such as work order quantity, return turns, user evaluation rate, return times of the robot and the like. The manager can perform basic mathematical operation self-defining indexes such as addition, subtraction, multiplication, division, power and the like on the basic data, and is used for establishing a service quality general index AN-PSR of the intelligent customer service robot, and the optimization strategy can optimize the service quality general index AN-PSR.
And the robot optimization suggestion module (3) is used for evaluating the performance of the intelligent customer service robot according to a set threshold and the index value of the intelligent customer service robot, and generating an optimization strategy to optimize the robot when the evaluation level of the intelligent customer service robot is lower than the set level.
9. The intelligent customer service robot data processing system of claim 8, further comprising:
the robot database module (6) is used for storing and calling related information such as a corpus of the daily robot, data of user question information, a robot training model and the like;
the user information understanding module (4) is used for translating the problems of the user into a character form by utilizing a voice recognition technology;
the semantic understanding module (7) is used for understanding the problems of the user, identifying the intentions of the problems of the user and outputting the content of the identification result;
and the data information display module (5) is used for displaying the key data and the administrator personalized data.
10. The intelligent customer service robot data processing system according to claim 8, wherein AN optimization strategy is generated when AN AN-PSR or custom evaluation level of the intelligent customer service robot is lower than a set level, the optimization strategy being capable of optimizing the robot database module (6) and the semantic understanding module (7).
CN202210974456.2A 2022-08-15 2022-08-15 Data processing method and system for intelligent customer service robot Pending CN115358774A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210974456.2A CN115358774A (en) 2022-08-15 2022-08-15 Data processing method and system for intelligent customer service robot

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210974456.2A CN115358774A (en) 2022-08-15 2022-08-15 Data processing method and system for intelligent customer service robot

Publications (1)

Publication Number Publication Date
CN115358774A true CN115358774A (en) 2022-11-18

Family

ID=84033779

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210974456.2A Pending CN115358774A (en) 2022-08-15 2022-08-15 Data processing method and system for intelligent customer service robot

Country Status (1)

Country Link
CN (1) CN115358774A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116596549A (en) * 2023-07-14 2023-08-15 山东交控科技有限公司 Question-answer response management method and system for rail transit customer service robot

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116596549A (en) * 2023-07-14 2023-08-15 山东交控科技有限公司 Question-answer response management method and system for rail transit customer service robot
CN116596549B (en) * 2023-07-14 2023-10-20 山东交控科技有限公司 Question-answer response management method and system for rail transit customer service robot

Similar Documents

Publication Publication Date Title
CN107329967B (en) Question answering system and method based on deep learning
CN108304372B (en) Entity extraction method and device, computer equipment and storage medium
CN111414479B (en) Label extraction method based on short text clustering technology
CN110033281B (en) Method and device for converting intelligent customer service into manual customer service
CN106991085B (en) Entity abbreviation generation method and device
CN110059182A (en) Art recommended method and device towards customer service
CN113535963B (en) Long text event extraction method and device, computer equipment and storage medium
CN110209790B (en) Question-answer matching method and device
US11429810B2 (en) Question answering method, terminal, and non-transitory computer readable storage medium
WO2003012685A2 (en) A data quality system
CN111309887B (en) Method and system for training text key content extraction model
CN111897930A (en) Automatic question answering method and system, intelligent device and storage medium
CN111797202B (en) Customer service conversation method based on artificial intelligence
CN110175273A (en) Text handling method, device, computer readable storage medium and computer equipment
CN110955767A (en) Algorithm and device for generating intention candidate set list set in robot dialogue system
CN113064980A (en) Intelligent question and answer method and device, computer equipment and storage medium
CN115878778A (en) Natural language understanding method facing business field
CN115358774A (en) Data processing method and system for intelligent customer service robot
CN117194647B (en) Intelligent question-answering system, method and device for offline environment
CN113836898A (en) Automatic order dispatching method for power system
CN116628173B (en) Intelligent customer service information generation system and method based on keyword extraction
CN112463944A (en) Retrieval type intelligent question-answering method and device based on multi-model fusion
CN115577080A (en) Question reply matching method, system, server and storage medium
CN111382265A (en) Search method, apparatus, device and medium
CN113111157A (en) Question-answer processing method, device, computer equipment and storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination