CN109993543A - A kind of complaint handling method and system - Google Patents
A kind of complaint handling method and system Download PDFInfo
- Publication number
- CN109993543A CN109993543A CN201711461567.9A CN201711461567A CN109993543A CN 109993543 A CN109993543 A CN 109993543A CN 201711461567 A CN201711461567 A CN 201711461567A CN 109993543 A CN109993543 A CN 109993543A
- Authority
- CN
- China
- Prior art keywords
- complaint
- decision tree
- reason
- entry information
- text entry
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000003066 decision tree Methods 0.000 claims abstract description 56
- 238000012549 training Methods 0.000 claims description 18
- 230000011664 signaling Effects 0.000 claims description 13
- 238000005516 engineering process Methods 0.000 claims description 9
- 238000004590 computer program Methods 0.000 claims description 8
- 238000013135 deep learning Methods 0.000 claims description 6
- 230000001052 transient effect Effects 0.000 claims description 6
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 239000000284 extract Substances 0.000 claims description 3
- 230000002265 prevention Effects 0.000 claims description 3
- 238000012512 characterization method Methods 0.000 claims description 2
- 238000013518 transcription Methods 0.000 abstract description 3
- 230000035897 transcription Effects 0.000 abstract description 3
- 238000013527 convolutional neural network Methods 0.000 description 15
- 230000008569 process Effects 0.000 description 15
- 238000012545 processing Methods 0.000 description 14
- 238000010586 diagram Methods 0.000 description 8
- 238000011176 pooling Methods 0.000 description 7
- 230000006870 function Effects 0.000 description 5
- 238000013138 pruning Methods 0.000 description 5
- 238000010276 construction Methods 0.000 description 3
- 235000019580 granularity Nutrition 0.000 description 3
- 235000012054 meals Nutrition 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000013519 translation Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 238000009795 derivation Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 239000002699 waste material Substances 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000013477 bayesian statistics method Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 230000008676 import Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 230000003449 preventive effect Effects 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 230000004083 survival effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000017105 transposition Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/01—Customer relationship services
- G06Q30/015—Providing customer assistance, e.g. assisting a customer within a business location or via helpdesk
- G06Q30/016—After-sales
-
- G—PHYSICS
- 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
- G10L15/26—Speech to text systems
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Databases & Information Systems (AREA)
- Human Computer Interaction (AREA)
- Artificial Intelligence (AREA)
- Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Finance (AREA)
- General Business, Economics & Management (AREA)
- Economics (AREA)
- Development Economics (AREA)
- Accounting & Taxation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Marketing (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Acoustics & Sound (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Health & Medical Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention provides a kind of complaint handling method, comprising: the complaint voice that will acquire is converted to the text entry information with index tab, and the index tab, which includes at least, complains classification;Based on the one-to-one decision tree of classification is complained, the highest complaint reason of number of votes obtained in the decision tree is determined, wherein each tree node of decision tree represents the characteristic attribute of a complaint reason;Based on the complaint reason, corresponding complaint solution is exported.A kind of complaint handling method and system provided by the invention, pass through artificial intelligent voice recognizer and decision Tree algorithms, the speech transcription that can be realized customer service hotline is structuring text, and it is accurately positioned according to word content combination decision Tree algorithms and complains reason, to which targetedly solution is complained in output, complaint handling efficiency is improved.
Description
Technical field
The present invention relates to field of communication service, more particularly, to a kind of complaint handling method and system.
Background technique
Under the network coexisted state of current 2/3/4G, disparate networks network element number of nodes is more, and networking structure is complicated, wirelessly
Under the influence of situations such as environment constantly deteriorates, user is discontented to mobile network quality so that the problem of complaining becomes increasingly conspicuous.
Complaint handling mode is in existing net at present: first by client service center personnel answer hot line receive all kinds of business applications,
Business is complained, network is complained etc., is formed with manual record and is complained work order;Pass through trouble ticket dispatch to core net, wireless, network management etc. again
Analytical derivation discovery, orientation problem are done by manually extracting related wireless index, network management index, signaling information etc. by relevant departments,
Landing solution is proposed according to all kinds of problems and is implemented, and follow up processing progress, eventually by customer service return visit client feedback problem
Whether solve.
But client service center answers all kinds of business such as the application of hot line processing business, customer complaint, substantial amounts and many and diverse,
Complain reason varied, customer service manual procedure very complicated inevitably generates incorrect posting, error of omission, cannot foundation to complaint reason
Customer complaint content carries out effective position by problem producing cause, and processing is caused to complain low efficiency and waste of manpower resource.
Summary of the invention
The present invention provides one kind and overcomes the problems, such as that above-mentioned processing complains low efficiency and waste of manpower resource or at least partly
A kind of complaint handling method that ground solves the above problems, comprising:
The complaint voice that will acquire is converted to the text entry information with index tab, and the index tab includes at least
Complain classification;
Based on the one-to-one decision tree of classification is complained, the highest complaint reason of number of votes obtained in the decision tree is determined,
In, each tree node of decision tree represents the characteristic attribute of a complaint reason;
Based on the complaint reason, corresponding complaint solution is exported.
A kind of complaint handling method and system provided by the invention are calculated by artificial intelligent voice recognizer and decision tree
Method, the speech transcription that can be realized customer service hotline is structuring text, and accurately fixed according to word content combination decision Tree algorithms
Reason is complained in position, so that targetedly solution is complained in output, improves complaint handling efficiency.
Detailed description of the invention
Fig. 1 is a kind of complaint handling method flow diagram provided in an embodiment of the present invention;
Fig. 2 is the decision tree schematic diagram provided in an embodiment of the present invention that the corresponding building of classification is complained about data service;
Fig. 3 is a kind of speech recognition process schematic diagram provided in an embodiment of the present invention;
Fig. 4 is complaint handling flow diagram provided in an embodiment of the present invention;
Fig. 5 is a kind of complaint handling system construction drawing provided in an embodiment of the present invention.
Specific embodiment
With reference to the accompanying drawings and examples, specific embodiments of the present invention will be described in further detail.Implement below
Example is not intended to limit the scope of the invention for illustrating the present invention.
In the prior art, due to the mode complained using artificial treatment, so that there are many problems, such as: customer service
All kinds of business such as the application of hot line processing business, customer complaint are answered at center, substantial amounts and many and diverse, complain reason varied,
Customer service manual procedure very complicated inevitably generates incorrect posting, error of omission, to complaining the reason cannot be according to customer complaint content by asking
Topic producing cause is effectively classified, and consumes a large amount of manpower and material resources and manual record forms and work order is complained to distribute inefficiency;
Trouble ticket dispatch exists and compared with too many levels leads to complaint handling poor in timeliness;Complaint problem link is handled in relevant departments, is needed artificial
It extracts the relevant wireless index of customer complaint, network management index, signaling etc. and does corresponding analysis discovery and orientation problem, treatment effeciency
It is low;It complains at present and lays particular emphasis on single-point complaint handling to the sudden complaint low-response of large area;Stress post-processing, preventive effect
Difference, customer satisfaction are low.
For above-mentioned problems of the prior art, the embodiment of the present invention provides a kind of complaint handling method and system,
The single, client service center by the application solution customer complaint channel such as intelligent semantic identification, big data algorithm, hadoop cluster technology
Manual record work order inefficiency spends human and material resources resource big, circulation work order poor in timeliness, complaint handling low efficiency etc. and asks
Topic.A set of energy is finally established for customer complaint efficient process, customer complaint can be predicted, can prevent, can be found prior to client
Problem solves network problem, and the complaint handling method of complaint amount can be effectively reduced.
Fig. 1 is a kind of complaint handling method flow diagram provided in an embodiment of the present invention, as shown in Figure 1, which comprises
S1, the complaint voice that will acquire are converted to the text entry information with index tab, and the index tab is at least
Including complaining classification;
S2, it is based on complaining the one-to-one decision tree of classification, determines the highest complaint reason of number of votes obtained in the decision tree,
Wherein, each tree node of decision tree represents the characteristic attribute of a complaint reason;
S3, it is based on the complaint reason, exports corresponding complaint solution.
It is understood that being taken about the complaint problem that mobile communication field is connected to often by client service center, wechat
The customer complaint voice of business number, complaint hotline etc. typing by all kinds of means.The prior art is to listen knowledge to obtain otherwise by manually debating
The content that voice is intended by, but often artificial treatment can not accomplish it is completely recorded, it is possible to some important informations can be omitted.
In embodiments of the present invention, intelligent semantic is used in S1 to know otherwise come at the voice data to acquisition
Reason can cross speech recognition technology and voice translate to the semantic parsing of progress in real time, when generation is after language person separation with Subscriber Number, complaint
Between, the text entry information that complaint location, complaint content are keyword, and by the text entry information according to above-mentioned keyword
Form be indexed storage, the corresponding voiced keyword of each index tab.
It is understood that will complain speech translation is structuring text information, voice is formed by this technological system
The structured database record that can be indexed, to create conditions to reach efficient automatic complaint processing.
The embodiment of the present invention has used decision Tree algorithms to position complaint in S2, finds problem Producing reason, from
And corresponding solution is exported in S3.
Positioning main method is delimited it is understood that complaining to solidify complaint handling method, this hair using known experience
The decision Tree algorithms of bright offer are a tree construction, each of which nonleaf node indicates the test on a characteristic attribute, Mei Gefen
Zhi represents output of this characteristic attribute in some codomain, and each leaf node stores a classification.It is carried out using decision tree
The process of decision is exactly to test corresponding characteristic attribute in item to be sorted, and according to the selection output point of its value since root node
Branch, until reaching leaf node, using the classification of leaf node storage as the result of decision.
The committed step of decision tree is Split Attribute.So-called Split Attribute is exactly at some node according to a certain feature category
The different demarcation of property constructs different branches, target be allow each division subset as much as possible " pure "." pure " is exactly as far as possible
Allow an oidiospore that item to be sorted is concentrated to belong to same category as far as possible.Split Attribute is divided into three kinds of different situations: attribute be from
It dissipates value and not seek survival into binary decision tree, use each division of attribute as a branch at this time;Attribute be discrete value and
It is required that generating binary decision tree.It is tested using a subset of Attribute transposition, according to " belonging to this subset " and " is not belonged at this time
In this subset " it is divided into Liang Ge branch;Attribute is successive value, determines that a value is used as split vertexes at this time, according to > node and≤
Node generates Liang Ge branch.
Fig. 2 is the decision tree schematic diagram provided in an embodiment of the present invention that the corresponding building of classification is complained about data service, such as
Shown in Fig. 2, according to the complaint classification of user, such as classification is complained in the data service provided in Fig. 2, can generate decision accordingly
Tree, votes on each tree node, in conjunction with subscriber signaling feature and through divided data feature, to each node of decision tree
Weight amendment is carried out, so that reason positioning result is accurately complained in output.
On the basis of the above embodiments, the method also includes:
Based on Taxonomy and distribution CART algorithm and CNN deep learning algorithm, prediction model is complained in building, and described in utilization
Text entry information with index tab is trained the complaint prediction model;
Based on the complaint prediction model after training, potential report user is exported, to carry out complaint prevention.
Described to be based on Taxonomy and distribution CART algorithm and CNN deep learning algorithm, prediction model is complained in building, and is utilized
The text entry information with index tab is trained the complaint prediction model, comprising:
Using signaling data and through divided data as the characterization factor of the CART algorithm and CNN deep learning algorithm, construct
The complaint prediction model;
History text record information, history signaling data and the history that will acquire are through divided data as training sample set pair
The complaint prediction model is trained.
It is understood that scheme provided in an embodiment of the present invention can not only effectively locate the complaint generated
Reason, can also to complain there may be the case where predict, to be prevented.
Specifically, use complaint prediction model based on big data in the embodiment of the present invention, using complaint historical data,
Signaling data passes through data cleansing, data quantization process through divided data etc., deep using big data method for digging CART algorithm, CNN
It spends learning algorithm and establishes complaint prediction model.
It should be noted that also utilizing complaint historical data, signaling number simultaneously during establishing and complaining prediction model
According to, be trained to model as training sample through divided data.
The CART is made of feature selecting, the generation of tree and beta pruning, and both can be used for classifying can be used for returning.
CART assumes that decision tree is binary tree, and the value of internal node feature is "Yes" and " no.Such decision tree is equivalent to recursively
Two points of each features.
Specific building complains the process of prediction model and training as follows:
Decision tree is generated based on training dataset, keeps the decision tree generated big as far as possible;
The generation of decision tree is exactly recursively to construct the process of binary decision tree.Regression tree square error is minimized quasi-
Then, criterion is minimized with gini index (Gini index) to classification tree, carries out feature selecting, generate binary tree.
Step 1 sets the training dataset of node as D, calculates existing feature to the Gini coefficient of the data set.At this point, right
D is divided into D1 for "Yes" or "No" to the test of A=a according to sample point by each feature A, each value a that may be taken to it
With D2 two parts, Gini coefficient when A=a is calculated.
Step 2, in all possible feature A and the possible cut-off a of all of which, select Gini coefficient it is the smallest
Feature and its corresponding cut-off are as optimal characteristics and optimal cut-off.According to optimal characteristics and optimal cut-off, from existing node
Two child nodes are generated, training dataset is assigned in two child nodes according to feature.
To two child nodes recursively invocation step l and step 2, until meeting preset stopping condition, to generate CART
Decision tree.
The condition that algorithm stops calculating is that the number of samples in node is less than predetermined threshold or the Gini coefficient of sample set is small
In predetermined threshold (sample substantially belongs to same class), or there is no more features.
In the embodiment of the present invention, the definition to Gini coefficient is in classification problem, it is assumed that has k class, sample point belongs to
The probability of kth class are as follows: pk, then the Gini Index Definition of probability distribution are as follows:
Table 1 as training sample set part through divided data
Arrearage | Preferential set meal >=3 | Whether complain |
It is | It is | It is no |
It is no | It is | It is no |
It is | It is no | It is |
It is | It is no | It is |
It is | It is no | It is |
It is | It is | It is |
It is | It is | It is |
It is no | It is no | It is no |
It is | It is | It is no |
It is no | It is | It is no |
Table 1 is the part as training sample set through divided data, as shown in table 1, through including throwing when adhering to separately property is " arrearage "
It tells classification 5, does not complain classification 2;Comprising complaining classification 0 when through adhering to separately property being " non-arrearage ", when not complaining classification 3:
Gini=1- [(5/7)2+(2/7)2]=0.41;
Gini=1- [(0/3)2+(3/7)2]=0.00.
Then Gini gain are as follows:
Gini (gain)=7/10* (1- (5/7)2-(2/7)2)+3/10*(1-(0/3)2-(3/7)2)=0.29.
Be through adhering to separately property " user preferential set meal number >=3 " when comprising complain class 2, do not complain class 4;It is through adhering to separately property "
When user preferential set meal number < 3 " comprising complaining class 1, not complaining class 3:
Gini=1- [(2/6)2+(4/6)2]=0.44
Gini=1- [(1/4)2+(3/4)2]=0.38
Then Gini gain are as follows:
Gini (gain)=6/10* (1- (2/6)2-(4/6)2)+4/10*(1-(1/4)2-(3/4)2)=0.42.
Thus the attribute of Gini (gain)=0.29 is selected to be divided for " arrearage ", according to such splitting rule CART
Algorithm can complete achievement process.
Further, beta pruning is carried out to generated tree with validation data set and selects optimal subtree, at this moment with loss letter
The minimum standard as beta pruning of number.CART pruning algorithms step are as follows:
The decision tree T generated first from generating algorithm0Low side starts continuous beta pruning, until T0Root node, form a son
Set sequence { T0,T1,…Tn}.Then subtree sequence is tested in independent validation data set by cross-validation method,
Therefrom choose optimal subtree.
The CNN deep learning algorithm specifically includes:
CNN convolutional neural networks are the neural network models of a multilayer, and most crucial place is convolution sum Pooling
Operation, convolution thought source is in the receptive field concept of human eye vision, i.e., a pocket centered on point of interest, and convolution more accords with
The essence for closing the two-dimensional space of image, can learn more effective feature;Pooling can simply be interpreted as down-sampled operation,
Learn the spatial feature of image.By the shared number that can reduce parameter of convolution kernel in CNN, the complexity of model is reduced.
It is some indeformable that Pooling can be such that the feature acquired has, such as translation, rotation, rotational invariance.
CNN training process is divided into 2 stages
Propagation stage forward: first stage takes a sample (X, Y) from sample set, X is inputted network;It calculates corresponding
Reality output O;In this stage, information, by transformation step by step, is transmitted to output layer from input layer.This process is also network
The process executed when being operated normally after completing training.In the process, network execute be calculate, actually input with
Every layer of weight matrix phase dot product, obtains output result to the end:
O=Fn (... (F2 (F1 (XW (1)) W (2)) ...) W (n)) second stage is the back-propagation stage: being calculated practical defeated
The difference of O and corresponding ideal output Y out;Weight matrix is adjusted by the method backpropagation of minimization error.
Second stage: CNN back-propagation algorithm
CNN cost function: it is typically chosen and minimizes square error (MSE) or minimum relative entropy (Relative
Entropy);Backpropagation generally uses stochastic gradient descent method.
The backpropagation of CNN mainly considers three aspects: output layer, the determination and derivation of cost function, similar BP network;
Pooling, the down-sampling of data and the up-sampling of residual error;Base, the convolution algorithm of data and the de-convolution operation of residual error are rolled up, such as
When next layer of fruit convolutional layer is pooling layers, need to do the up-sampling of residual error.If Pooling is using maxpooling's
Words need the position for recording maximum value in propagated forward to need to do residual error if next layer of Pooling is convolutional layer
Deconvolution.
CNN convolutional neural networks network is the Hierarchical Neural Networks of multilayer, and every layer of neuron is all same type, or
Simply, or complicated or hypercomplex neuron, there is very rare and fixed mode connection between each layer there.When need
The feature wanted has predefined, then network learns layer by layer, space between hidden layer and hidden layer with regard to using there is supervision algorithm
Resolution ratio is successively decreased, and number of planes contained by every layer is incremented by, and can be used for detecting more characteristic informations in this way, thus constantly this
A little characteristic informations are updated into complaint handling, prediction, prevention module.
It is understood that complaint problem can be learnt using the CNN machine depth self-learning algorithm based on artificial intelligence
Processing optimizes complaint handling process, realizes complaint handling progress chasing, and timely feedback result to user, makes complaint handling shape
At effective closed-loop.
Shown in sum up, prediction mould is complained in features, the buildings such as the embodiment of the present invention passes through big data report user signaling, divided
Type exports potential report user, solves network problem prior to customer complaint, active inquiry is carried out to such user, to prevent
It complains.
On the basis of the above embodiments, the complaint voice that will acquire is converted to the text entry with index tab
Information, the index tab, which includes at least, complains classification, comprising:
Based on speech recognition technology, the complaint voice that will acquire is converted to text entry information;
It extracts the keyword in the text entry information and is stored according to the corresponding index tab of the keyword,
To obtain the text entry information with index tab.
Described to be based on speech recognition technology, the complaint voice that will acquire is converted to the text entry information with keyword,
Include:
By in the speech recognition modeling after the complaint voice input training, it is special to extract the corresponding voice of the complaint voice
Sign;
Search is decoded to the phonetic feature, the complaint voice is converted into text entry information.
Specifically, speech recognition technology provided in an embodiment of the present invention is based on Bayesian statistics modeling framework, (MAP/ is maximum
Posterior probability decision rule), Plug-In MAP.
Wherein, acoustic feature X is extracted by front end features and is obtained, acoustic modelAcoustic feature is counted
Modeling, language modelStatistical modeling is carried out to word string, decoding search --- optimal word string is obtained by algorithm for design
On this basis, speech recognition technology is based on feed-forward type serial memorization network (Feed-forward Sequential
Memory Network, FSMN) new frame, be by one kind based on DNN modified network structure.Draw in the hidden layer of DNN
Enter time-delay structure, using the hidden layer historical information at t-N~t-1 moment as next layer of input, to introduce voice sequence
Historical information, while avoid RNN training BPTT bring gradient disappear, the problems such as computation complexity is high, further increase
Speech recognition accuracy.
Fig. 3 is a kind of speech recognition process schematic diagram provided in an embodiment of the present invention, as shown in figure 3, the embodiment of the present invention
It can be by the voice data largely marked training acoustic model, to pass through acoustic model and language model for the voice outside training set
Data are identified as text.
Further, when being received again by by hotline of complaint, special line, the complaint of wechat public platform voice, voice is complained defeated
After entering intelligent sound identification translation system, structured index is carried out to text after transcription, when being stored as with Subscriber Number, complaint
Between, the database record information that complaint location, complaint content etc. are keyword, and to complaining hot spot to carry out automatic cluster.
On the basis of the above embodiments, described based on the corresponding decision tree of classification is complained, it determines in the decision tree and obtains
The highest complaint reason of poll, comprising:
Based on the corresponding relationship between the complaint classification and the decision tree, objective decision tree is determined;
The text entry information with index tab and External Data Representation XDR signaling data are inputted into the target
In decision tree, the highest complaint reason of number of votes obtained in the objective decision tree is determined.
Fig. 4 is complaint handling flow diagram provided in an embodiment of the present invention, as shown in figure 4, by with index tab
In the decision tree that text entry information and the input of External Data Representation XDR signaling data have been built up, complaint reason is determined
Position exports corresponding solution according to localized reason.
On the basis of the above embodiments, the index tab includes complaint location, the method also includes:
Count all text entry information with index tab converted in preset time period;
Based on the complaint location in the index tab, real-time display complains hot spot region in Distribution GIS.
It is understood that the embodiment of the present invention is to apply hadoop cluster framework to carry out height to the relevant data of complaint
The XDR data received can be pushed to hadoop in such a way that TXT text formatting is by socket by speed processing, the embodiment of the present invention
Cluster.Hadoop is received using storm and converting system pushes the TXT text file to come, is then reused by caching, data
HDFS is issued by certain mechanism after compression to be stored.Hive and Spark mainly to the file that HDFS store carry out inquiry and
Processing, the former focuses on non-real-time data query processing, and the latter focuses on real-time data query process.After query processing
As a result ORACLE database purchase can be imported, and foreground application is developed based on ORACLE.
TXT file that is that Storm is received and being stored in HDFS be each interface XDR (imsi, imei, msisdn,
Failure_cause, request_cause, user_ipv4, user_ipv6 etc.).Hadoop cluster can to the above XDR into
The corresponding Data Integration of row, generates various customized XDR, and be conducted into ORACLE database, facilitates application server tune
With.
It is understood that in embodiments of the present invention, hadoop cluster mainly completes data query function, wherein wrapping
It includes: to the inquiry of data real-time statistics is complained, the XDR text text being stored in HDFS is purposefully inquired using Hive function
Part, and query result directly displayed, exports or imported into ORACLE with formats such as TXT/CSV.
Based on XDR text file, can it is customized, automatically generate all kinds of convergence Table X DR, and result is imported
ORACLE complains hot spot region day granularity convergence table (complain_day) for example, can generate based on complain table.
Or using Spark function efficiently, rapidly inquire XDR text file, generate the convergence of various small time granularities
Table X DR, such as 5 minutes or 15 minutes, and it is imported into ORACLE, heat is complained for example, can generate based on complain_min table
15 minutes granularities in point region import ORACLE, and application program converges tables by these of calling ORACLE, and real-time statistics are presented
The case where customer complaint hot spot region.
It can also be presented in real time in GIS map, it is to be understood that the embodiment of the present invention can collect some period
All complaints inside received judge to complain hot spot region and show that can also predict according to prediction result real-time display may
The complaint region of generation, facilitates contact staff to handle immediately.
Further, the embodiment of the present invention is also provided is generated by complaining demarcation locating module to output results to application server
Failure/network problem handles work order, and auto form delivering gives relevant departments' processing, and timely feedbacks processing progress, eventually by customer service,
The various ways such as wechat, short message pay a return visit whether client feedback problem solves.
Fig. 5 is a kind of complaint handling system construction drawing provided in an embodiment of the present invention, as shown in figure 5, the system comprises:
Voice conversion module 1 complains reason locating module 2 and complains solution module 3, in which:
The complaint voice that voice conversion module 1 is used to will acquire is converted to the text entry information with index tab, institute
It states index tab and includes at least complaint classification;
Reason locating module 2 is complained to be used to determine number of votes obtained in the decision tree based on the corresponding decision tree of classification is complained
Highest complaint reason;
It complains and solves module 3 for being based on the complaint reason, output is corresponding to complain solution.
It is specific how to utilize 3 pairs of module voice conversion module 1, complaint reason locating module 2 and complaint solution complaints
Carrying out processing can be found in above-described embodiment, and details are not described herein for the embodiment of the present invention.
The embodiment of the present invention provides a kind of complaint handling system, comprising: at least one processor;And with the processor
At least one processor of communication connection, in which:
The memory is stored with the program instruction that can be executed by the processor, and the processor calls described program to refer to
It enables to execute method provided by above-mentioned each method embodiment, for example, the complaint voice that will acquire, which is converted to, has index
The text entry information of label, the index tab, which includes at least, complains classification;Based on the corresponding decision tree of classification is complained, determine
The highest complaint reason of number of votes obtained in the decision tree;Based on the complaint reason, corresponding complaint solution is exported.
The present embodiment discloses a kind of computer program product, and the computer program product includes being stored in non-transient calculating
Computer program on machine readable storage medium storing program for executing, the computer program include program instruction, when described program instruction is calculated
When machine executes, computer is able to carry out method provided by above-mentioned each method embodiment, for example, the complaint voice that will acquire
The text entry information with index tab is converted to, the index tab, which includes at least, complains classification;Based on complaint classification pair
The decision tree answered determines the highest complaint reason of number of votes obtained in the decision tree;Based on the complaint reason, corresponding throw is exported
Tell solution.
The present embodiment provides a kind of non-transient computer readable storage medium, the non-transient computer readable storage medium
Computer instruction is stored, the computer instruction makes the computer execute method provided by above-mentioned each method embodiment, example
It such as include: that the complaint voice that will acquire is converted to the text entry information with index tab, the index tab includes at least
Complain classification;Based on the corresponding decision tree of classification is complained, the highest complaint reason of number of votes obtained in the decision tree is determined;Based on institute
Complaint reason is stated, corresponding complaint solution is exported.
Those of ordinary skill in the art will appreciate that: realize that all or part of the steps of above method embodiment can pass through
The relevant hardware of program instruction is completed, and program above-mentioned can be stored in a computer readable storage medium, the program
When being executed, step including the steps of the foregoing method embodiments is executed;And storage medium above-mentioned includes: ROM, RAM, magnetic disk or light
The various media that can store program code such as disk.
Through the above description of the embodiments, those skilled in the art can be understood that each embodiment can
It realizes by means of software and necessary general hardware platform, naturally it is also possible to pass through hardware.Based on this understanding, on
Stating technical solution, substantially the part that contributes to existing technology can be embodied in the form of software products in other words, should
Computer software product may be stored in a computer readable storage medium, such as ROM/RAM, magnetic disk, CD, including several fingers
It enables and using so that a computer equipment (can be personal computer, server or the network equipment etc.) executes each implementation
Method described in certain parts of example or embodiment.
Finally, the present processes are only preferable embodiment, it is not intended to limit the scope of the present invention.It is all
Within the spirit and principles in the present invention, any modification, equivalent replacement, improvement and so on should be included in protection of the invention
Within the scope of.
Claims (10)
1. a kind of complaint handling method characterized by comprising
The complaint voice that will acquire is converted to the text entry information with index tab, and the index tab, which includes at least, complains
Classification;
Based on the one-to-one decision tree of classification is complained, the highest complaint reason of number of votes obtained in the decision tree is determined, wherein certainly
Each tree node of plan tree represents the characteristic attribute of a complaint reason;
Based on the complaint reason, corresponding complaint solution is exported.
2. the method according to claim 1, wherein the method also includes:
Based on Taxonomy and distribution CART algorithm and CNN deep learning algorithm, prediction model is complained in building, and is had described in utilization
The text entry information of index tab is trained the complaint prediction model;
Based on the complaint prediction model after training, potential report user is exported, to carry out complaint prevention.
3. according to the method described in claim 2, it is characterized in that, described deep based on Taxonomy and distribution CART algorithm and CNN
Learning algorithm, building complaint prediction model are spent, and is complained in advance using the text entry information with index tab to described
Model is surveyed to be trained, comprising:
Using signaling data and through divided data as the characterization factor of the CART algorithm and CNN deep learning algorithm, described in building
Complain prediction model;
History text record information, history signaling data and the history that will acquire are through divided data as training sample set to described
Prediction model is complained to be trained.
4. the method according to claim 1, wherein the complaint voice that will acquire, which is converted to, has index mark
The text entry information of label, the index tab, which includes at least, complains classification, comprising:
Based on speech recognition technology, the complaint voice that will acquire is converted to text entry information;
It extracts the keyword in the text entry information and is stored according to the corresponding index tab of the keyword, with
To the text entry information for having index tab.
5. according to the method described in claim 4, it is characterized in that, described be based on speech recognition technology, the complaint language that will acquire
Sound is converted to the text entry information with keyword, comprising:
By in the speech recognition modeling after the complaint voice input training, the corresponding phonetic feature of the complaint voice is extracted;
Search is decoded to the phonetic feature, the complaint voice is converted into text entry information.
6. the method according to claim 1, wherein described be based on complaining the one-to-one decision tree of classification, really
The highest complaint reason of number of votes obtained in the fixed decision tree, wherein each tree node of decision tree represents a complaint reason
Characteristic attribute, comprising:
Based on the corresponding relationship between the complaint classification and the decision tree, objective decision tree is determined;
The text entry information with index tab and External Data Representation XDR signaling data are inputted into the objective decision
In tree, the highest complaint reason of number of votes obtained in the objective decision tree is determined.
7. the method is also the method according to claim 1, wherein the index tab includes complaint location
Include:
Count all text entry information with index tab converted in preset time period;
Based on the complaint location in the index tab, real-time display complains hot spot region in Distribution GIS.
8. a kind of complaint handling system characterized by comprising
Voice conversion module, the complaint voice for will acquire are converted to the text entry information with index tab, the rope
Tendering label, which include at least, complains classification;
Reason locating module is complained, for determining number of votes obtained in the decision tree based on the one-to-one decision tree of classification is complained
Highest complaint reason, wherein each tree node of decision tree represents the characteristic attribute of a complaint reason;
It complains and solves module, for being based on the complaint reason, output is corresponding to complain solution.
9. a kind of computer program product, which is characterized in that the computer program product includes being stored in non-transient computer
Computer program on readable storage medium storing program for executing, the computer program include program instruction, when described program is instructed by computer
When execution, the computer is made to execute the method as described in claim 1 to 7 is any.
10. a kind of non-transient computer readable storage medium, which is characterized in that the non-transient computer readable storage medium is deposited
Computer instruction is stored up, the computer instruction makes the computer execute the method as described in claim 1 to 7 is any.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711461567.9A CN109993543A (en) | 2017-12-28 | 2017-12-28 | A kind of complaint handling method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711461567.9A CN109993543A (en) | 2017-12-28 | 2017-12-28 | A kind of complaint handling method and system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109993543A true CN109993543A (en) | 2019-07-09 |
Family
ID=67108356
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711461567.9A Pending CN109993543A (en) | 2017-12-28 | 2017-12-28 | A kind of complaint handling method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109993543A (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110363419A (en) * | 2019-07-10 | 2019-10-22 | 拉扎斯网络科技(上海)有限公司 | Task data processing method and device, electronic equipment and medium |
CN110516748A (en) * | 2019-08-29 | 2019-11-29 | 泰康保险集团股份有限公司 | Method for processing business, device, medium and electronic equipment |
CN110766586A (en) * | 2019-10-22 | 2020-02-07 | 恒瑞通(福建)信息技术有限公司 | Method and device for processing environmental complaint events |
CN111429157A (en) * | 2020-03-27 | 2020-07-17 | 上海东普信息科技有限公司 | Method, device and equipment for evaluating and processing complaint work order and storage medium |
CN112188014A (en) * | 2020-10-28 | 2021-01-05 | 中国联合网络通信集团有限公司 | Information processing method and device |
CN113343711A (en) * | 2021-06-29 | 2021-09-03 | 南方电网数字电网研究院有限公司 | Work order generation method, device, equipment and storage medium |
CN113810212A (en) * | 2020-06-15 | 2021-12-17 | ***通信集团浙江有限公司 | Root cause positioning method and device for 5G slice user complaints |
CN115442832A (en) * | 2021-06-03 | 2022-12-06 | ***通信集团四川有限公司 | Complaint problem positioning method and device and electronic equipment |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102419975A (en) * | 2010-09-27 | 2012-04-18 | 深圳市腾讯计算机***有限公司 | Data mining method and system based on voice recognition |
CN107293285A (en) * | 2016-04-06 | 2017-10-24 | ***通信集团四川有限公司 | A kind of method of speech processing and device |
KR20170140714A (en) * | 2016-06-13 | 2017-12-21 | 안드로메다(주) | Advertisement output system using the call waiting time |
-
2017
- 2017-12-28 CN CN201711461567.9A patent/CN109993543A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102419975A (en) * | 2010-09-27 | 2012-04-18 | 深圳市腾讯计算机***有限公司 | Data mining method and system based on voice recognition |
CN107293285A (en) * | 2016-04-06 | 2017-10-24 | ***通信集团四川有限公司 | A kind of method of speech processing and device |
KR20170140714A (en) * | 2016-06-13 | 2017-12-21 | 안드로메다(주) | Advertisement output system using the call waiting time |
Non-Patent Citations (1)
Title |
---|
刘世伟等: "基于客户投诉信息的创新预测方法研究 ", 《移动通信》 * |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110363419A (en) * | 2019-07-10 | 2019-10-22 | 拉扎斯网络科技(上海)有限公司 | Task data processing method and device, electronic equipment and medium |
CN110516748A (en) * | 2019-08-29 | 2019-11-29 | 泰康保险集团股份有限公司 | Method for processing business, device, medium and electronic equipment |
CN110766586A (en) * | 2019-10-22 | 2020-02-07 | 恒瑞通(福建)信息技术有限公司 | Method and device for processing environmental complaint events |
CN111429157A (en) * | 2020-03-27 | 2020-07-17 | 上海东普信息科技有限公司 | Method, device and equipment for evaluating and processing complaint work order and storage medium |
CN113810212A (en) * | 2020-06-15 | 2021-12-17 | ***通信集团浙江有限公司 | Root cause positioning method and device for 5G slice user complaints |
CN112188014A (en) * | 2020-10-28 | 2021-01-05 | 中国联合网络通信集团有限公司 | Information processing method and device |
CN112188014B (en) * | 2020-10-28 | 2022-04-26 | 中国联合网络通信集团有限公司 | Information processing method and device |
CN115442832A (en) * | 2021-06-03 | 2022-12-06 | ***通信集团四川有限公司 | Complaint problem positioning method and device and electronic equipment |
CN115442832B (en) * | 2021-06-03 | 2024-04-09 | ***通信集团四川有限公司 | Complaint problem positioning method and device and electronic equipment |
CN113343711A (en) * | 2021-06-29 | 2021-09-03 | 南方电网数字电网研究院有限公司 | Work order generation method, device, equipment and storage medium |
CN113343711B (en) * | 2021-06-29 | 2024-05-10 | 南方电网数字电网研究院有限公司 | Work order generation method, device, equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109993543A (en) | A kind of complaint handling method and system | |
CN105808590B (en) | Search engine implementation method, searching method and device | |
CN110033342A (en) | A kind of training method and device, a kind of recommended method and device of recommended models | |
US11804216B2 (en) | Generating training datasets for a supervised learning topic model from outputs of a discovery topic model | |
JP2020521210A (en) | Information processing method and terminal, computer storage medium | |
CN113949582B (en) | Network asset identification method and device, electronic equipment and storage medium | |
CN109493186A (en) | The method and apparatus for determining pushed information | |
CN114265921A (en) | Question-answer knowledge base construction method and device, equipment, medium and product thereof | |
US20230004587A1 (en) | Bot for query analysis and interface generation | |
CN108959327B (en) | Service processing method, device and computer readable storage medium | |
CN113111157B (en) | Question-answer processing method, device, computer equipment and storage medium | |
CN111241826B (en) | Entity name recognition method, device, equipment and storage medium | |
CN114492452A (en) | Method, device and equipment for training and appealing switching of pre-training language model | |
KR20230023505A (en) | Method and apparatus for generating spoken language understanding model | |
CN112380332A (en) | Interactive knowledge feedback method, device and computer storage medium | |
CN112287084A (en) | Question-answering method and system based on ensemble learning | |
CN117473070B (en) | Multi-channel application method of intelligent robot, intelligent robot and storage medium | |
CN117714603B (en) | Outbound method, outbound device and readable storage medium | |
Nuntalid et al. | Poster: Machine Learning for Predicting Emergency Medical Incidents that Need an Air-ambulance | |
CN110442692A (en) | It is a kind of for problem worksheet processing and its method and apparatus of training | |
CN113672711B (en) | Service type robot intention recognition device and training and recognition method thereof | |
CN116776870B (en) | Intention recognition method, device, computer equipment and medium | |
CN113676603B (en) | Call control method, device, equipment and storage medium | |
CN116541506B (en) | Intelligent dialogue method, device, equipment and storage medium based on machine learning | |
CN117834780B (en) | Intelligent outbound customer intention prediction analysis system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190709 |