CN109471922A - Case type recognition methods, device, equipment and medium based on deep learning model - Google Patents
Case type recognition methods, device, equipment and medium based on deep learning model Download PDFInfo
- Publication number
- CN109471922A CN109471922A CN201811148701.4A CN201811148701A CN109471922A CN 109471922 A CN109471922 A CN 109471922A CN 201811148701 A CN201811148701 A CN 201811148701A CN 109471922 A CN109471922 A CN 109471922A
- Authority
- CN
- China
- Prior art keywords
- case
- deep learning
- learning model
- keyword
- type
- 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
Abstract
The invention discloses a kind of case type recognition methods based on deep learning model, device, computer equipment and storage medium, obtain the target case of case type to be distinguished and the communication information of client's expression, then information input to deep learning model will be linked up and obtain the first input results as case keyword, because deep learning model is linked up information and history case keyword as sample training by history and is obtained, and trained deep learning model has efficient operating capability in advance, so case keyword rapidly can be filtered out from communication information, when case keyword includes preset complaint keyword, and information is linked up when not including preset consulting keyword, the case type for determining target case is complaining type, when case keyword does not include preset complaint keyword, or case keyword includes preset official communication When asking keyword, the case type of target case is determined for consulting type, so improving the efficiency of case type judgement.
Description
Technical field
The present invention relates to deep learning technology field more particularly to it is a kind of based on deep learning model based on deep learning
Case type recognition methods, device, computer equipment and the storage medium of model.
Background technique
Currently, insurance company is provided with consulting for preferably services client, exclusively for client in insurance financial industry
With complaint service, such as consulting and complaint service hotline.
With the growth of consulting and the caseload complained, within a preset period, many clients are using same
When a kind of mode is seeked advice from or complained, such as Please ring for assistance or complains, and contact staff needs high concentration energy to accept a large amount of visitor
The case of consulting or complaint that family is initiated, due to the energy of contact staff or mood etc., contact staff occurs sometimes can not
The case type for rapidly differentiating case belongs to the case where seeking advice from type or complaining type, the efficiency for causing case type to judge
Lowly.
Therefore, finding one kind, efficiently the case type recognition methods based on deep learning model becomes those skilled in the art
The problem of member's urgent need to resolve.
Summary of the invention
The embodiment of the present invention provides a kind of to be set based on the case type recognition methods of deep learning model, device, computer
Standby and storage medium, to solve the problems, such as the inefficiency of case type judgement.
A kind of case type recognition methods based on deep learning model, comprising:
Obtain the target case of case type to be distinguished；
Obtain the communication information that client expresses in the target case；
The communication information input to preparatory trained deep learning model is obtained into the first output result as case
Keyword, wherein the deep learning model is to link up information and history target keywords as sample training by history to obtain；
When the case keyword includes preset complaint keyword, and the case keyword does not include preset consulting
When keyword, determine that the case type of the target case is complaining type；
When the case keyword does not include preset complaint keyword or the case keyword includes preset consulting
When keyword, determine the case type of the target case for consulting type.
A kind of case type identification device based on deep learning model, comprising:
Obtain the target case of case type to be distinguished；
Obtain the communication information that client expresses in the target case；
The communication information input to preparatory trained deep learning model is obtained into the first output result as case
Keyword, wherein the deep learning model is to link up information and history target keywords as sample training by history to obtain；
When the case keyword includes preset complaint keyword, and the case keyword does not include preset consulting
When keyword, determine that the case type of the target case is complaining type；
When the case keyword does not include preset complaint keyword or the case keyword includes preset consulting
When keyword, determine the case type of the target case for consulting type.
A kind of computer equipment, including memory, processor and storage are in the memory and can be in the processing
The computer program run on device, the processor are realized above-mentioned based on deep learning model when executing the computer program
The step of case type recognition methods.
A kind of computer readable storage medium, the computer-readable recording medium storage have computer program, the meter
Calculation machine program realizes the step of above-mentioned case type recognition methods based on deep learning model when being executed by processor.
Above-mentioned case type recognition methods, device, computer equipment and storage medium based on deep learning model, passes through
It obtains the target case of case type to be distinguished and obtains the communication information of client's expression in the target case, it then will be described
It links up information input to preparatory trained deep learning model and obtains the first input results as case keyword, because of depth
Learning model is linked up information and history target keywords as sample training by history and is obtained, and trained deep learning in advance
Model has efficient operating capability, so case keyword rapidly can be filtered out from the communication information, when described
Case keyword includes preset complaint keyword, and when the communication information does not include preset consulting keyword, determines institute
The case type for stating target case is complaining type, when the case keyword does not include preset complaint keyword or described
When case keyword includes preset consulting keyword, the case type of the target case is determined for consulting type, so mentioning
The high efficiency of case type judgement.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, below by institute in the description to the embodiment of the present invention
Attached drawing to be used is needed to be briefly described, it should be apparent that, the accompanying drawings in the following description is only some implementations of the invention
Example, for those of ordinary skill in the art, without any creative labor, can also be according to these attached drawings
Obtain other attached drawings.
Fig. 1 is that an application environment of the case type recognition methods in one embodiment of the invention based on deep learning model is shown
It is intended to；
Fig. 2 is a flow chart of the case type recognition methods in one embodiment of the invention based on deep learning model；
Fig. 3 is training deep learning in case type recognition methods in one embodiment of the invention based on deep learning model
One flow chart of model；
Fig. 4 be in one embodiment of the invention based in the case type recognition methods of deep learning model determine negative sample and
One flow chart of positive sample；
Fig. 5 is to obtain customer voice in case type recognition methods in one embodiment of the invention based on deep learning model
One flow chart of information；
Fig. 6 is a schematic diagram of the case type identification device in one embodiment of the invention based on deep learning model；
Fig. 7 is a schematic diagram of computer equipment in one embodiment of the invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are some of the embodiments of the present invention, instead of all the embodiments.Based on this hair
Embodiment in bright, every other implementation obtained by those of ordinary skill in the art without making creative efforts
Example, shall fall within the protection scope of the present invention.
Case type recognition methods provided by the present application based on deep learning model can be applicable to and apply ring such as Fig. 1
In border, wherein computer equipment is communicated by network with server.Wherein, computer equipment can be, but not limited to various
Personal computer, laptop, smart phone, tablet computer and portable wearable device.Server can be with independent
The server cluster of server either multiple servers composition is realized.
It in one embodiment, should as shown in Fig. 2, providing a kind of case type recognition methods based on deep learning model
Case type recognition methods based on deep learning model is applied in financial industry, applies the service in Fig. 1 in this way
It is illustrated, includes the following steps: for device
S10, the target case for obtaining case type to be distinguished；
In embodiments of the present invention, case type includes complaining type and consulting type.The target case generation of complaining type
Table client complains, and the target case for seeking advice from type represents client and seeks advice from.
Specifically, the event usually exchanged with client is referred to as target case, which is stored in advance in
It in a case part database, can be called at any time when needing, obtain storage of the target case in First Case part database first
Then the target case of case type to be distinguished, such as customer service are extracted in path according to the store path from First Case part database
The event that personnel are exchanged with " Zhang San " is target case, and First Case part database is SQL database, then first obtains the target
Store path of the case in SQL database, then according to the store path extract the contact staff of case type to be distinguished with
The target case that " Zhang San " is exchanged.
It should be noted that First Case part database can be SQL database or orable database, case database
Particular content can be set, herein with no restrictions according to actual needs.
S20, the communication information for obtaining client's expression in target case；
In embodiments of the present invention, the communication information of client's expression has voice or text.When linking up information is voice, visitor
Family can link up information using phone or the phonetic function of social software, such as the phonetic function of wechat to carry out expression；Work as ditch
When communication breath is text, client can using sending short messages, writing letter or the text input function of social software, if mobile phone is sent short messages,
Or wechat dispatch word etc..Client expresses the particular content for linking up the expression way of information, can be set according to practical application,
Herein with no restrictions.
Specifically, the communication information that client's expression is obtained from target case, the communication information expressed such as client are that " I wants
Your company is complained to Insurance Regulation office ".
S30, communication information input to preparatory trained deep learning model is obtained into the first output result as case
Keyword；
In embodiments of the present invention, trained deep learning model can be convolutional neural networks model in advance, may be used also
Think other models, the particular content of preparatory trained deep learning model can be set, herein according to practical application
With no restrictions.
Wherein, which is to link up information and history mesh by the history of client's expression in history target case
Mark keyword is obtained as sample training.
Specifically, the communication information input for the client's expression that will acquire to preparatory trained deep learning model obtains
First output is as a result, and using the first obtained output result as case keyword.
Judge whether case keyword includes preset complaint keyword, and whether case keyword does not include preset official communication
Keyword is ask, if so, S40 is thened follow the steps, if it is not, thening follow the steps S50.
S40, the case type for determining target case are complaining type；
In embodiments of the present invention, preset complaint keyword can be " I will complain " or " your uncle " etc., it is preset
Seek advice from keyword can be " hello, I will seek advice from down " or " may I ask, if under capable of seek advice from " etc., preset complaint keyword in advance
If consulting keyword particular content, can be set according to practical application, herein with no restrictions.
Specifically, when obtained case keyword includes preset complaint keyword, and obtained case keyword is not
When including preset consulting keyword, just determine that the case type of the target case is complaining type, namely represent client's progress
It complains, the customer complaint is that client leads throwing to media, network, government department, Industry regulatory bodies or company executives
It tells.
It is further possible to be recorded a video using preset remote-recording video equipment to the behavior act of client, then sentence
Whether the behavior act of disconnected client is preset complaint movement, which is the behavior that client makes in indignation
Desk is such as patted in movement, if so, the case type for determining the target case is complaining type, namely is represented client and is thrown
It tells, if it is not, determining the case type of the target case just to seek advice from type, namely represents client and seek advice from.
It should be noted that preset remote-recording video equipment can be digital camera or video recorder etc., preset long-range record
As the particular content that equipment and preset complaint act, set according to actual needs, herein with no restrictions.
S50, determine the case type of target case for consulting type.
Specifically, when obtained case keyword does not include preset complaint keyword group, or obtained case keyword
When including preset consulting keyword group, just determine the case type of the target case for consulting type, namely represent client into
Row consulting.
In the corresponding embodiment of Fig. 2, by first obtaining the target case of case type to be distinguished, goal-trail is then obtained
Next the communication information that client expresses in part will link up information input to preparatory trained deep learning model and obtain first
Result is exported as case keyword, wherein deep learning model is to link up information and history target keywords conduct by history
Sample training obtains, and next when case keyword includes preset complaint keyword, and case keyword does not include presetting
Consulting keyword when, determine target case case type be complaining type, finally when case keyword does not include preset
When complaint keyword or case keyword include preset consulting keyword, determine the case type of target case for consulting class
Type.By obtaining the target case of case type to be distinguished and obtaining the communication information of client's expression in the target case, so
The communication information input to preparatory trained deep learning model is obtained into the first input results as case keyword afterwards,
Because deep learning model is linked up information and history target keywords as sample training by history and is obtained, and trained in advance
Deep learning model has efficient operating capability, so case key can be filtered out rapidly from the communication information
Word, when the case keyword includes preset complaint keyword, and the communication information does not include preset consulting keyword
When, determine that the case type of the target case is complaining type, when the case keyword does not include that preset complaint is crucial
When word or the case keyword include preset consulting keyword, determine the case type of the target case for consulting class
Type, so improving the efficiency of case type judgement.
In one embodiment, the case type recognition methods based on deep learning model is somebody's turn to do to apply in big data treatment industry
In, training depth in a kind of case type recognition methods based on deep learning model in Fig. 2 corresponding embodiment as shown in Figure 3
Flow chart of the model under an application scenarios is practised, is specifically comprised the following steps:
S301, the history for obtaining client's expression in history case link up information and history target keywords as sample；
Specifically, for specially storing history case, acquisition history case first exists usually the second case database of setting
Then store path in second case database extracts the history case according to the store path from the second case database
Part, the history that next client expresses from obtaining in history case link up information, next obtain history target keywords,
The history that client expresses in the history case that will acquire links up information and the history target keywords got are sample.
It should be noted that the second database can be SQL database or orable database, case database it is specific
Content can be set, herein with no restrictions according to actual needs.
S302, the history in sample is linked up into information input to the deep learning model, obtains the second output result；
Specifically, the history in the sample that will acquire links up information input to the deep learning model, by depth
After the analysis of learning model, the second output result is obtained.
The hidden layer parameter of S303, the adjustment deep learning model, to minimize the second output result and going through in sample
The error of history target keywords；
In embodiments of the present invention, hidden layer parameter includes neurode number, improved stepping constant, target are accurate every time
Rate, maximum number of iterations and cost function.
Specifically, using the second output result as output target, the hidden layer parameter of the deep learning model is adjusted, thus
Reach minimizing the error between the history target keywords in the second output result and sample.
It should be noted that in adjusting parameter, by first adjusting stepping constant, the i.e. speed of observation cost function decline
Rate corrects stepping constant, on the other hand one side rapid decrease prevents from not restraining, and then, adjusts again after adjustment is suitable hidden
Node layer number to be hidden, is gradually increased, accuracy rate should first be increased, it is rear to reduce, after finding suitable interstitial content,
Finally, target accuracy rate is gradually turned up.
Further, judge whether the error of the second output result and the history target keywords in sample meets default item
Part, if so, S304 is thened follow the steps, if it is not, S301 to step S303 is then returned to step, until the error meets default item
Until part.
It should be noted that preset condition can be 0.01% or 0.015%, the particular content of preset condition can be with
It is set according to practical application, herein with no restrictions.
S304, determine that current deep learning model is trained deep learning model.
Specifically, when the error of the history target keywords in the second output result and sample meets preset condition, really
Deep learning model before settled is trained deep learning model；When the history target in the second output result and sample is closed
When the error of key word is unsatisfactory for preset condition, determine that current deep learning model is not trained deep learning model.
In the corresponding embodiment of Fig. 3, information and history are linked up by first obtaining the history that client expresses in history case
Then history in sample is linked up information input to the deep learning model, obtains second by target keywords as sample
Output adjusts the hidden layer parameter of the deep learning model as a result, next using the second output result as target, to minimize
The error of second output result and the history target keywords in sample, if error meets preset condition, it is determined that current
Deep learning model is trained deep learning model.Due to using a large amount of historical data as sample, historical data packet
It includes the history that client expresses in history case and links up information and history target keywords, while these historical datas are by true
History case is got, and is linked up information as input using the history in sample and put into deep learning model, exported as a result,
Then, the output target of result percentage regulation learning model the most will be exported, is constantly communicated using authentic and valid history ditch
Breath and history target keywords are adjusted the hidden layer parameter of adjustment deep learning model, so that it is guaranteed that deep learning model
Output result is minimized the error with the history target keywords in sample, to ensure that the case of deep learning model output
The accuracy of keyword.
In one embodiment, the case type recognition methods based on deep learning model is somebody's turn to do to apply in big data treatment industry
In, negative sample is determined in Fig. 2 corresponding embodiment as shown in Figure 4 in a kind of case type recognition methods based on deep learning model
With flow chart of the positive sample under an application scenarios, specifically comprise the following steps:
S60, judge case keyword whether be non-punctuation mark text；
Specifically, due to the output result of deep learning model may for punctuation mark or escape character etc., such as "? " or
"/", not necessarily text, need to judge case keyword whether be non-punctuation mark text.
If S70, case keyword are not the texts of non-punctuation mark, information will be linked up and the determination of case keyword is negative
Sample, negative sample is for updating deep learning model；
Specifically, when case keyword is not the text of non-punctuation mark, information will be linked up and case keyword determines
For negative sample, so that updating deep learning model using negative sample.
If S80, case keyword are the texts of non-punctuation mark, it is determined that case keyword is to determine that obtained case is closed
Key word, and information and case keyword will be linked up and be determined as positive sample, positive sample is for updating deep learning model.
Specifically, when case keyword is the text of non-punctuation mark, determine that case keyword is to determine obtained case
Part keyword, and information and case keyword will be linked up and be determined as positive sample, so that updating deep learning using positive sample
Model.
In the corresponding embodiment of Fig. 4, by first judge case keyword whether be non-punctuation mark text, if then
Case keyword is not the text of non-punctuation mark, then will link up information and case keyword is determined as negative sample, negative sample is used
In updating deep learning model, if last case keyword is the text of non-punctuation mark, it is determined that case keyword is to determine
Obtained case keyword, and history communication information and case keyword are determined as positive sample, positive sample is for updating depth
Learning model.Since the output result of deep learning model may be punctuation mark or escape character etc., need to judge that case is closed
Whether key word is that the text of non-punctuation mark will link up information and case when case keyword is not the text of non-punctuation mark
Part keyword is determined as negative sample, when case keyword is the text of non-punctuation mark, determines that case keyword is determining
The case keyword arrived, and history communication information and case keyword are determined as positive sample, it thereby may be ensured that deep learning
While model analysis obtains output result, the effect of training amendment error also can achieve, and then improve deep learning mould
The accuracy rate of type analysis ability.
In one embodiment, the case type recognition methods based on deep learning model is somebody's turn to do to apply in big data treatment industry
In, client's language is obtained in Fig. 2 corresponding embodiment as shown in Figure 5 in a kind of case type recognition methods based on deep learning model
Message ceases the flow chart under an application scenarios, specifically comprises the following steps:
S201, during contact staff communicates with client, start specified recording system and record to the voice of client,
It obtains linking up voice document；
In embodiments of the present invention, specifying recording system includes specifying main recording system and specified priming system for electrical teaching.
Specifically, it during contact staff communicates with client, is recorded first according to the setting judgement in system by dynamic
Sound module carries out dynamic record and still carries out static recording by static recording module.Dynamic is carried out according to dynamic record module
Recording, then start dynamic record module, and then monitoring whether there is sound in verbal system, if so, starting the language to client
Sound is recorded, and voice document is obtained；In Recording Process, by the mute duration set in mute duration module and recorded
Time in journey is compared, and when mute time, which is greater than, sets mute duration, then communication close link, stops recording, and adopt
Periodic sound monitoring is carried out to verbal system with mute duration is set.When mute time be less than set mute duration when, then after
Continuous recording.During the mute duration of the use setting carries out periodic sound monitoring to verbal system, if being sent out within the kth period
Existing sound occurs, then main recording system restores k-1 cycle data, records since the kth period.Dynamic record module uses language
The mode of sound excitation is recorded, and can be recorded according to the speech of client, and the automatic empty information for intercepting not sound is broken traditional
Static recording mechanism reduces the time of editor's recording to provide important audio document for feedback opinion analysis.
It should be noted that k and k-1 is positive integer.
S202, monitoring specify whether main recording system works normally, when specifying main recording system to be in non-normal working shape
When state, starts specified priming system for electrical teaching and record to voice, when specifying main recording system to be in normal operating conditions, then open
It is dynamic that main recording system is specified to record voice, simultaneously close off specified priming system for electrical teaching；
Specifically, in order to preferably record, guarantee going on smoothly for Recording Process, whether just to need to monitor main recording system
Often work, when main recording system is in abnormal operating state, starting priming system for electrical teaching records to call, when main recording
When system is in normal operating conditions, starting specifies main recording system to record voice, while then closing specified priming sound
System.When carrying out recording work, main recording system can record to verbal system using network access system, the network
It can be wifi, 3G, 4G or GPRS etc., priming system for electrical teaching can use digital trunk to call by traditional doubling mode
Equipment is recorded.
It should be noted that the particular content of network, can be set, herein with no restrictions according to practical application.
S203, speech recognition is carried out to voice document is linked up using preset speech recognition tools, obtains linking up text work
To link up information.
Specifically, speech recognition is carried out to voice document is linked up using preset speech recognition tools, obtains linking up text,
And using obtained communication text as communication information.
Further, the summation of the text in the communication information be greater than preset first threshold when, e.g., voice document when
Long field Time is accurate to the second, and the corresponding text number Number field of voice document is int (integer), works as Number/Time
It is too fast to be determined as client's word speed greater than 10/3, it is determined that the case type of the target case is complaining type.
It should be noted that preset speech recognition tools can be Baidu's speech engine or Iflytek speech engine,
The particular content of preset speech recognition tools can be set, herein with no restrictions according to practical application.It is described to pass through hundred
It spends speech engine and carries out speech recognition to voice document is linked up, obtain linking up text, successively include syntactic analysis, big data point
Analysis, feature extraction, pattern match, arithmetic analysis and result screen and etc. rear input recognition result, it is thus understood that, Baidu's voice
The identification process of engine is the prior art, is described in detail in the embodiment of the present invention not to this.
In the corresponding embodiment of Fig. 5, by first starting specified recording system during contact staff communicates with client
It records to the voice of client, obtains linking up voice document, wherein specified recording system includes specifying main recording system and referring to
Determine priming system for electrical teaching, then monitoring specifies whether main recording system works normally, when specifying main recording system to be in improper work
When making state, starts specified priming system for electrical teaching and record to voice, when specifying main recording system to be in normal operating conditions,
Then starting specifies main recording system to record voice, simultaneously closes off specified priming system for electrical teaching, finally uses preset voice
Identification facility carries out speech recognition to voice document is linked up, and obtains linking up text as communication information.The voice to client into
When row recording, if sound signal quality is good, it can be recorded by main recording system, so that recording effect is higher, letter
Cease transmission speed it is very fast, can very fast handling failure can then be carried out by priming system for electrical teaching when sound signal quality is bad
Recording guarantees that recorded message is complete, guarantees the integrality of recorded message transmitting to improve recording quality.
In one embodiment, the case type recognition methods based on deep learning model is somebody's turn to do to apply in big data treatment industry
In, after the step s 40, being somebody's turn to do the case type recognition methods based on deep learning model further includes following steps:
S90, use preset sending method that case type is sent to full-time processing people for the target case of complaining type
Member, so that full-time treatment people completes the processing of target case.
Specifically, in order to preferably services client, after the case type that target case has been determined is complaining type,
Use preset sending method that case type is sent to full-time treatment people for the target case of complaining type, so that full-time
The processing for the treatment of people completion target case.
Further, priority level can also be arranged in the target case that case type is complaining type, priority level is got over
The urgency of the target case of high complaining type is higher, so that full-time treatment people is according to the height of the urgency of target case
Sequential processes mark case.
Further, the emergency rating that the target case mark of different priority levels can also be given different, works as target
When the priority level of case is greater than or equal to preset second threshold, indicate for red early warning state, it is preferential when target case
Rank is greater than or equal to preset third threshold value, and when being less than preset second threshold, indicates for blue alert status, work as target
When the priority level of case is less than preset third threshold value, indicate white alert status.
It should be noted that preset sending method can be for by sending short message, sending wechat or sending Mail Contents
Etc. sending methods, the particular content of preset sending method can be set according to practical application, herein with no restrictions.
In this embodiment, the target case that case type is complaining type is sent out by using preset sending method
It send to full-time treatment people, so that full-time treatment people completes the processing of target case at the first time, to improve processing
The timeliness of target case.
It should be understood that the size of the serial number of each step is not meant that the order of the execution order in above-described embodiment, each process
Execution sequence should be determined by its function and internal logic, the implementation process without coping with the embodiment of the present invention constitutes any limit
It is fixed.
In one embodiment, a kind of case type identification device based on deep learning model is provided, depth should be based on
Practise model case type identification device and above-described embodiment in the case type recognition methods based on deep learning model one by one
It is corresponding.It is obtained as shown in fig. 6, being somebody's turn to do the case type identification device based on deep learning model including the first acquisition module 601, second
Modulus block 602, the first input module 603, the first determining module 604 and the second determining module 605.Each functional module is described in detail
It is as follows:
First obtains module 601, for obtaining the target case of case type to be distinguished；
Second obtains module 602, for obtaining the communication information that client expresses in target case；
First input module 603 obtains first for that will link up information input to preparatory trained deep learning model
Result is exported as case keyword, wherein deep learning model is to link up information and history target keywords conduct by history
Sample training obtains；
First determining module 604, for including preset complaint keyword when case keyword, and case keyword does not wrap
When including preset consulting keyword, determine that the case type of target case is complaining type；
Second determining module 605, for not including preset complaint keyword or case keyword packet when case keyword
When including preset consulting keyword, determine the case type of target case for consulting type.
Further, it is somebody's turn to do the case type identification device based on deep learning model further include:
Third obtains module, links up information and history target keywords for obtaining the history that client expresses in history case
As sample；
Second input module obtains for the history in sample to be linked up information input to the deep learning model
Two output results；
Module is adjusted, for adjusting the hidden layer parameter of the deep learning model, to minimize the second output result and sample
The error of history target keywords in this；
Third determining module, if meeting preset condition for error, it is determined that current deep learning model is training
Good deep learning model.
Further, it is somebody's turn to do the case type identification device based on deep learning model further include:
Judgment module, for judge case keyword whether be non-punctuation mark text；
4th determining module will link up information and case if for case keyword not being the text of non-punctuation mark
Keyword is determined as negative sample, and negative sample is for updating deep learning model；
5th determining module, if for case key word being the text of non-punctuation mark, it is determined that case keyword is to determine
Obtained case keyword, and information and case keyword will be linked up and be determined as positive sample, positive sample is for updating deep learning
Model.
Further, linking up information is voice, and the second acquisition module 602 includes:
Recording module, for during contact staff communicates with client, starting specified recording system to the voice of client
It records, obtains linking up voice document, wherein specified recording system includes specifying main recording system and specified priming phonetic system
System；
Monitoring modular specifies whether main recording system works normally for monitoring, when specify main recording system be in it is non-just
When normal working condition, start specified priming system for electrical teaching and record to voice, works normally shape when specifying main recording system to be in
When state, then starting specifies main recording system to record voice, simultaneously closes off specified priming system for electrical teaching；
Identification module obtains ditch for carrying out speech recognition to voice document is linked up using preset speech recognition tools
Logical text is as communication information.
Further, it is somebody's turn to do the case type identification device based on deep learning model further include:
Sending module, for using preset sending method to be sent to case type specially for the target case of complaining type
Duty treatment people, so that full-time treatment people completes the processing of target case.
Specific restriction about the case type identification device based on deep learning model may refer to above for base
In the restriction of the case type recognition methods of deep learning model, details are not described herein.The above-mentioned case based on deep learning model
Modules in part type identification device can be realized fully or partially through software, hardware and combinations thereof.Above-mentioned each module
It can be embedded in the form of hardware or independently of in the processor in computer equipment, computer can also be stored in a software form
In memory in equipment, the corresponding operation of the above modules is executed in order to which processor calls.
In one embodiment, a kind of computer equipment is provided, which can be server, internal junction
Composition can be as shown in Figure 7.The computer equipment include by system bus connect processor, memory, network interface and
Database.Wherein, the processor of the computer equipment is for providing calculating and control ability.The memory packet of the computer equipment
Include non-volatile memory medium, built-in storage.The non-volatile memory medium is stored with operating system, computer program and data
Library.The built-in storage provides environment for the operation of operating system and computer program in non-volatile memory medium.The calculating
The database of machine equipment is for storing the data information that the case type recognition methods based on deep learning model is related to.The meter
The network interface for calculating machine equipment is used to communicate with external terminal by network connection.When the computer program is executed by processor
To realize a kind of case type recognition methods based on deep learning model.
In one embodiment, a kind of computer equipment is provided, including memory, processor and storage are on a memory
And the computer program that can be run on a processor, processor realize that above-described embodiment is based on depth when executing computer program
The step of practising the case type recognition methods of model, such as step S10 shown in Fig. 2 to step S50.Alternatively, processor executes
Each module/unit of the case type identification device in above-described embodiment based on deep learning model is realized when computer program
Function, such as the function of the first acquisition 601 to the second determining module 605 of module shown in Fig. 6.It is no longer superfluous here to avoid repeating
It states.
In one embodiment, a kind of computer readable storage medium is provided, computer program is stored thereon with, is calculated
The case type recognition methods based on deep learning model in above method embodiment is realized when machine program is executed by processor, or
Person realizes that the case type in above-mentioned apparatus embodiment based on deep learning model is known when the computer program is executed by processor
The function of each module/unit in other device.To avoid repeating, which is not described herein again.Those of ordinary skill in the art will appreciate that
It realizes all or part of the process in above-described embodiment method, is that can instruct relevant hardware come complete by computer program
At the computer program can be stored in a non-volatile computer read/write memory medium, which is holding
When row, it may include such as the process of the embodiment of above-mentioned each method.Wherein, used in each embodiment provided herein pair
Memory, storage, any reference of database or other media, may each comprise non-volatile and/or volatile memory.It is non-easy
The property lost memory may include that read-only memory (ROM), programming ROM (PROM), electrically programmable ROM (EPROM), electric erasable can
Programming ROM (EEPROM) or flash memory.Volatile memory may include that random access memory (RAM) or external speed buffering are deposited
Reservoir.By way of illustration and not limitation, RAM is available in many forms, such as static state RAM (SRAM), dynamic ram (DRAM), synchronizes
DRAM (SDRAM), double data rate sdram (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronization link (Synchlink)
DRAM (SLDRAM), memory bus (Rambus) directly RAM (RDRAM), direct memory bus dynamic ram (DRDRAM),
And memory bus dynamic ram (RDRAM) etc..
It is apparent to those skilled in the art that for convenience of description and succinctly, only with above-mentioned each function
Can unit, module division progress for example, in practical application, can according to need and by above-mentioned function distribution by different
Functional unit, module are completed, i.e., the internal structure of described device is divided into different functional unit or module, more than completing
The all or part of function of description.
Embodiment described above is merely illustrative of the technical solution of the present invention, rather than its limitations；Although referring to aforementioned reality
Applying example, invention is explained in detail, those skilled in the art should understand that: it still can be to aforementioned each
Technical solution documented by embodiment is modified or equivalent replacement of some of the technical features；And these are modified
Or replacement, the spirit and scope for technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution should all
It is included within protection scope of the present invention.
Claims (10)
1. a kind of case type recognition methods based on deep learning model, which is characterized in that described to be based on deep learning model
Case type recognition methods include:
Obtain the target case of case type to be distinguished；
Obtain the communication information that client expresses in the target case；
The communication information input to preparatory trained deep learning model is obtained into the first output result as case key
Word, wherein the deep learning model is to link up information and history target keywords as sample training by history to obtain；
When the case keyword includes preset complaint keyword, and the case keyword does not include that preset consulting is crucial
When word, determine that the case type of the target case is complaining type；
When the case keyword does not include preset complaint keyword or the case keyword includes that preset consulting is crucial
When word, determine the case type of the target case for consulting type.
2. as described in claim 1 based on the case type recognition methods of deep learning model, which is characterized in that the depth
Learning model is to link up information and history target keywords as sample training by history to obtain, the training deep learning model
It specifically includes:
The history for obtaining client's expression in history tip-offs about environmental issues links up information and history target keywords as sample；
History in the sample is linked up into information input to the deep learning model, obtains the second output result；
The hidden layer parameter of the deep learning model is adjusted, to minimize the history in the second output result and the sample
Error between target keywords；
If the error meets preset condition, it is determined that the current deep learning model is trained deep learning mould
Type.
3. as described in claim 1 based on the case type recognition methods of deep learning model, which is characterized in that will be described
Information input is linked up to after trained deep learning model obtains the first output result as case keyword in advance, it is described
Case type recognition methods based on deep learning model further include:
Judge the case keyword whether be non-punctuation mark text；
If the case keyword is not the text of non-punctuation mark, the communication information and the case keyword are determined
For negative sample, the negative sample is for updating the deep learning model；
If the case keyword is the text of non-punctuation mark, it is determined that the case keyword is to determine the obtained case
Part keyword, and the communication information and the case keyword are determined as positive sample, the positive sample is described for updating
Deep learning model.
4. as described in claim 1 based on the case type recognition methods of deep learning model, which is characterized in that the communication
Information is voice, and the communication information for obtaining client's expression in the target case includes:
During contact staff communicates with client, starts specified recording system and record to the voice of client, obtain described
Link up voice document, wherein the specified recording system includes specifying main recording system and specified priming system for electrical teaching；
Monitor it is described specify whether main recording system works normally, specify main recording system to be in abnormal operating state when described
When, start the specified priming system for electrical teaching and record to the voice, specifies main recording system to be in normal work when described
When state, then specifies main recording system to record the voice using described, simultaneously close off specified priming system for electrical teaching；
Speech recognition is carried out to the communication voice document using preset speech recognition tools, obtains linking up described in text conduct
Link up information.
5. according to any one of claims 1 to 4 based on the case type recognition methods of deep learning model, feature exists
In, the determination target case case type be complaining type after, the case based on deep learning model
Kind identification method further include:
Use preset sending method that case type is sent to full-time treatment people for the target case of complaining type, so that
The sole duty treatment people completes the processing of target case.
6. a kind of case type judgment means, which is characterized in that the case type identification device based on deep learning model
Include:
First obtains module, for obtaining the target case of case type to be distinguished；
Second obtains module, for obtaining the communication information that client expresses in the target case；
First input module, for the communication information input to preparatory trained deep learning model to be obtained the first output
As a result it is used as case keyword, wherein the deep learning model is to link up information and history target keywords conduct by history
Sample training obtains；
First determining module, for including preset complaint keyword when the case keyword, and the case keyword is not
When including preset consulting keyword, determine that the case type of the target case is complaining type；
Second determining module, for not including preset complaint keyword or the case keyword when the case keyword
When including preset consulting keyword, determine the case type of the target case for consulting type.
7. as claimed in claim 6 based on the case type identification device of deep learning model, which is characterized in that described to be based on
The case type identification device of deep learning model further include:
Third obtains module, links up information and history target keywords for obtaining the history that client expresses in history tip-offs about environmental issues
As sample；
Second input module obtains for the history in the sample to be linked up information input to the deep learning model
Two output results；
Module is adjusted, for adjusting the hidden layer parameter of the deep learning model, to minimize the second output result and institute
State the error of the history target keywords in sample；
Third determining module, if meeting preset condition for the error, it is determined that the current deep learning model is
Trained deep learning model.
8. the case type identification device based on deep learning model as described in any one of claim 6 to 7, feature exist
In the case type identification device based on deep learning model further include:
Judgment module, for judge the case keyword whether be non-punctuation mark text；
4th determining module, if for the case keyword not being the text of non-punctuation mark, by the communication information and
The case keyword is determined as negative sample, and the negative sample is for updating the deep learning model；
5th determining module, if for the case keyword being the text of non-punctuation mark, it is determined that the case keyword
To determine the obtained case keyword, and the communication information and the case keyword are determined as positive sample, it is described
Positive sample is for updating the deep learning model.
9. a kind of computer equipment, including memory, processor and storage are in the memory and can be in the processor
The computer program of upper operation, which is characterized in that the processor realized when executing the computer program as claim 1 to
The step of case type recognition methods described in any one of 5 based on deep learning model.
10. a kind of computer readable storage medium, the computer-readable recording medium storage has computer program, and feature exists
In realization is based on deep learning model as described in any one of claims 1 to 5 when the computer program is executed by processor
Case type recognition methods the step of.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811148701.4A CN109471922A (en) | 2018-09-29 | 2018-09-29 | Case type recognition methods, device, equipment and medium based on deep learning model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811148701.4A CN109471922A (en) | 2018-09-29 | 2018-09-29 | Case type recognition methods, device, equipment and medium based on deep learning model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109471922A true CN109471922A (en) | 2019-03-15 |
Family
ID=65664837
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811148701.4A Pending CN109471922A (en) | 2018-09-29 | 2018-09-29 | Case type recognition methods, device, equipment and medium based on deep learning model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109471922A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110458082A (en) * | 2019-08-05 | 2019-11-15 | 城云科技（中国）有限公司 | A kind of city management case classification recognition methods |
| CN110490098A (en) * | 2019-07-31 | 2019-11-22 | 恒大智慧科技有限公司 | Smoking behavior automatic testing method, equipment and the readable storage medium storing program for executing of community user |
| CN111241274A (en) * | 2019-12-31 | 2020-06-05 | 航天信息股份有限公司 | Criminal law document processing method and device, storage medium and electronic device |
| CN111522947A (en) * | 2020-04-22 | 2020-08-11 | 北京思特奇信息技术股份有限公司 | Method and system for processing complaint work order |
| CN112214925A (en) * | 2020-09-17 | 2021-01-12 | 上海微亿智造科技有限公司 | Sintering process product quality characteristic processing method and system suitable for deep model training |
| CN112910754A (en) * | 2020-05-07 | 2021-06-04 | 腾讯科技（深圳）有限公司 | Message processing method, device, equipment and storage medium based on group session |
| CN113222443A (en) * | 2021-05-25 | 2021-08-06 | 支付宝(杭州)信息技术有限公司 | Case shunting method and device |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018032937A1 (en) * | 2016-08-19 | 2018-02-22 | 中兴通讯股份有限公司 | Method and apparatus for classifying text information |
| CN108351890A (en) * | 2015-11-24 | 2018-07-31 | 三星电子株式会社 | Electronic device and its operating method |
| CN108416375A (en) * | 2018-02-13 | 2018-08-17 | 中国联合网络通信集团有限公司 | Work order sorting technique and device |
| CN108491386A (en) * | 2018-03-19 | 2018-09-04 | 上海携程国际旅行社有限公司 | natural language understanding method and system |
| US20180261213A1 (en) * | 2017-03-13 | 2018-09-13 | Baidu Usa Llc | Convolutional recurrent neural networks for small-footprint keyword spotting |
-
2018
- 2018-09-29 CN CN201811148701.4A patent/CN109471922A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108351890A (en) * | 2015-11-24 | 2018-07-31 | 三星电子株式会社 | Electronic device and its operating method |
| WO2018032937A1 (en) * | 2016-08-19 | 2018-02-22 | 中兴通讯股份有限公司 | Method and apparatus for classifying text information |
| US20180261213A1 (en) * | 2017-03-13 | 2018-09-13 | Baidu Usa Llc | Convolutional recurrent neural networks for small-footprint keyword spotting |
| CN108416375A (en) * | 2018-02-13 | 2018-08-17 | 中国联合网络通信集团有限公司 | Work order sorting technique and device |
| CN108491386A (en) * | 2018-03-19 | 2018-09-04 | 上海携程国际旅行社有限公司 | natural language understanding method and system |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110490098A (en) * | 2019-07-31 | 2019-11-22 | 恒大智慧科技有限公司 | Smoking behavior automatic testing method, equipment and the readable storage medium storing program for executing of community user |
| CN110458082A (en) * | 2019-08-05 | 2019-11-15 | 城云科技（中国）有限公司 | A kind of city management case classification recognition methods |
| CN110458082B (en) * | 2019-08-05 | 2022-05-03 | 城云科技（中国）有限公司 | Urban management case classification and identification method |
| CN111241274A (en) * | 2019-12-31 | 2020-06-05 | 航天信息股份有限公司 | Criminal law document processing method and device, storage medium and electronic device |
| CN111241274B (en) * | 2019-12-31 | 2023-11-28 | 航天信息股份有限公司 | Criminal legal document processing method and device, storage medium and electronic equipment |
| CN111522947A (en) * | 2020-04-22 | 2020-08-11 | 北京思特奇信息技术股份有限公司 | Method and system for processing complaint work order |
| CN112910754A (en) * | 2020-05-07 | 2021-06-04 | 腾讯科技（深圳）有限公司 | Message processing method, device, equipment and storage medium based on group session |
| CN112214925A (en) * | 2020-09-17 | 2021-01-12 | 上海微亿智造科技有限公司 | Sintering process product quality characteristic processing method and system suitable for deep model training |
| CN113222443A (en) * | 2021-05-25 | 2021-08-06 | 支付宝(杭州)信息技术有限公司 | Case shunting method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109471922A (en) | Case type recognition methods, device, equipment and medium based on deep learning model | |
| EP3525438B1 (en) | Artificial intelligence based service implementation | |
| US11956187B2 (en) | Natural language processing for information extraction | |
| US11128680B2 (en) | AI mediated conference monitoring and document generation | |
| US10057419B2 (en) | Intelligent call screening | |
| CN112492111B (en) | Intelligent voice outbound method, device, computer equipment and storage medium | |
| CN111212190B (en) | Conversation management method, device and system based on conversation strategy management | |
| CN110351443A (en) | Intelligent outgoing call processing method, device, computer equipment and storage medium | |
| US11893545B2 (en) | Mobile secretary cloud application | |
| CN110956956A (en) | Voice recognition method and device based on policy rules | |
| CN108073600A (en) | A kind of intelligent answer exchange method, device and electronic equipment | |
| CN107610695A (en) | Driver's voice wakes up the dynamic adjusting method of instruction word weight | |
| US20160117624A1 (en) | Intelligent meeting enhancement system | |
| CN109688281A (en) | A kind of intelligent sound exchange method and system | |
| CN110493019B (en) | Automatic generation method, device, equipment and storage medium of conference summary | |
| CN103916513A (en) | Method and device for recording communication message at communication terminal | |
| CN109977202A (en) | A kind of intelligent customer service system and its control method | |
| US11811962B2 (en) | Mobile secretary cloud application | |
| CN109614627A (en) | A kind of text punctuate prediction technique, device, computer equipment and storage medium | |
| CN116049360A (en) | Intelligent voice dialogue scene conversation intervention method and system based on client image | |
| CN107392781A (en) | The recognition methods of customer relationship, the recognition methods of object relationship and device | |
| CN112925888A (en) | Method and device for training question-answer response and small sample text matching model | |
| CN111611365A (en) | Flow control method, device, equipment and storage medium of dialog system | |
| CN109120613B (en) | Communication data monitoring method and system | |
| US20230007123A1 (en) | Method and apparatus for automated quality management of communication records |
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 |