CN110472008A

CN110472008A - Intelligent interactive method and device

Info

Publication number: CN110472008A
Application number: CN201910600784.4A
Authority: CN
Inventors: 陈鑫
Original assignee: Alibaba Group Holding Ltd
Current assignee: Advanced New Technologies Co Ltd; Advantageous New Technologies Co Ltd
Priority date: 2019-07-04
Filing date: 2019-07-04
Publication date: 2019-11-19
Anticipated expiration: 2039-07-04
Also published as: CN110472008B

Abstract

This specification provides intelligent interactive method and device, wherein the intelligent interactive method includes: the interaction request for receiving user, the interactive instruction between the user and the target object is carried in the interaction request；By the learning algorithm model trained in advance of the interactive instruction input between the user and the target object；Adapted information relevant with the target object corresponding to the interactive instruction is exported by the learning algorithm model；The Adapted information is sent to the user.This specification exports corresponding Adapted information, is conducive to the validity for improving interaction, while being also beneficial to improve the accuracy of the Adapted information by the learning algorithm model that the interactive instruction input in interaction request is trained in advance.

Description

Intelligent interactive method and device

Technical field

This specification is related to robotic technology field, in particular to a kind of intelligent interactive method.This specification is related to simultaneously A kind of intelligent interaction device, a kind of electronic equipment and a kind of computer readable storage medium.

Background technique

With the rapid development of economy, user group is higher and higher to the attention rate of insurance, the number insured is also increasingly It is more, therefore in order to allow user more fast and easily to complete to insure, different from buying insurance model, many lines under traditional wire Service platform of insuring comes into being therewith.

The marketing method of insurance is more and more diversified, but no matter user is insured or passed through by the insurance company under online Line service platform of insuring is insured, and user is both needed to understand insurance by certain channel before formally insuring, and understands oneself institute The insurance kind needed, now in the art, due to the development of internet, numerous insurance companies are realized to a certain extent by by user The voice messaging of input carries out speech recognition or is quickly identified the text information of user's typing, is existed according to recognition result Answer content needed for positioning user in knowledge base.

But in the prior art, this mode of answer content needed for positioning user in knowledge base according to recognition result is more The professional knowledge richness of the integrity degrees for relying on knowledge base and agent itself are carrying out intelligent query information and recommendation more Accuracy of information in terms of there is also certain defects.

Summary of the invention

In view of this, this specification embodiment provides a kind of intelligent interactive method.This specification is related to a kind of intelligence simultaneously Energy interactive device, a kind of electronic equipment and a kind of computer readable storage medium, to solve technology existing in the prior art Defect.

According to this specification embodiment in a first aspect, providing a kind of intelligent interactive method, comprising:

The interaction request of user is received, the interaction between the user and the target object is carried in the interaction request Instruction；

By the learning algorithm model trained in advance of the interactive instruction input between the user and the target object；

It relevant with the target object is adapted to by the way that learning algorithm model output is corresponding to the interactive instruction Information；

The Adapted information is sent to the user.

Optionally, the learning algorithm model is trained in the following manner:

Obtain the administrative mechanism information and the target pair of customer attribute information, target object attribute information, target object As relevant rudimentary knowledge information and case data relevant to the target object；

With customer attribute information, target object attribute information, the administrative mechanism information of target object and the target object Relevant rudimentary knowledge information and case data relevant to the target object are training sample training learning algorithm model.

Optionally, after the interaction request step for receiving user executes, further includes:

The identification information of the user is obtained according to default identification algorithm；

If determining that the user is target user according to the identification information, study prompt is sent to the user Information；

The target user is detected for the response results of the study prompt information；

If detect the response results for confirmation study, by system interaction interface be the target user show with The relevant information of the target object.

Optionally, after the Adapted information step execution to user transmission, further includes:

According to preset condition collect interactive instruction between the user and the target object and with the interactive instruction pair The Adapted information answered；

By between the user and the target object interactive instruction and Adapted information corresponding with the interactive instruction add The training sample is added to form new training sample, the learning algorithm model is carried out based on the new training sample Model optimization.

Interactive instruction between the user and the target object, corresponding with the interactive instruction is collected according to preset condition Adapted information and user to the feedback score result of the Adapted information corresponding with the interactive instruction；

By feedback score result be higher than preset fraction threshold value the user and the target object between interactive instruction and Adapted information corresponding with the interactive instruction is added to the training sample to form new training sample, based on described new Training sample carries out model optimization to the learning algorithm model.

Optionally, the interactive instruction is voice question information；

After the interaction request step for receiving user executes, further includes:

Speech recognition is carried out to obtain corresponding text information to the voice question information；

The learning algorithm model packet that interactive instruction input between the user and the target object is trained in advance It includes:

By text information input learning algorithm model trained in advance.

Optionally, before the learning algorithm model sub-step that text information input is trained in advance executes, also Include:

Semantic analysis is carried out to the text information, obtains corresponding semantic analysis result；

The learning algorithm model that text information input is trained in advance includes:

By semantic analysis result input learning algorithm model trained in advance.

Keyword extraction is carried out to the text information according to preset rules；

By the keyword input of extraction learning algorithm model trained in advance.

Optionally, the intelligent interactive method, further includes:

The uniform resource location of Initial page is obtained according to preset rules；

Link and by institute relevant to the target object is extracted in the Initial page by web page analysis algorithm State the acquisition for linking and being added to uniform resource location queue outstanding message relevant to the target object；

Judge whether that reaching the default information task that obtains stops execution condition, if it is not, then repeating according to preset rules The step of obtaining the uniform resource location of Initial page；

If so, the information to acquisition is handled, and data store to treated.

Optionally, after the information of described pair of acquisition carries out processing sub-step execution, further includes:

By treated, content is added to the training sample to form new training sample, based on the new training sample This carries out model optimization to the learning algorithm model.

Optionally, described that and the target object corresponding with the interactive instruction is exported by the learning algorithm model After relevant Adapted information step executes, Xiang Suoshu user is sent before the Adapted information step execution, further includes:

The Adapted information is analyzed, determines the corresponding language form of the Adapted information；

Based on the speech samples library and the language form prestored, generated and the Adapted information by text transformation technology Corresponding voice messaging；

It is described to include: to the user transmission Adapted information

Play the voice messaging.

According to the another aspect of this specification embodiment, a kind of intelligent interaction device is provided, comprising:

Interaction request receiving module is configured as receiving the interaction request of user, carry in the interaction request described Interactive instruction between user and the target object；

Interactive instruction input module is configured as the interactive instruction input between the user and the target object is preparatory Trained learning algorithm model；

Adapted information output module is configured as exporting by the learning algorithm model corresponding with the interactive instruction Adapted information relevant to the target object；

Adapted information sending module is configured as sending the Adapted information to the user.

Optionally, the intelligent interaction device, further includes:

Module is obtained, is configured as obtaining the administrative mechanism of customer attribute information, target object attribute information, target object Information, rudimentary knowledge information relevant to the target object and case data relevant with the target object；

Model training module, be configured as with customer attribute information, target object attribute information, target object supervisor Information, rudimentary knowledge information relevant to the target object and case data relevant with the target object processed are training Sample training learning algorithm model.

Optionally, the intelligent interaction device, further includes:

Identity information acquisition module is configured as obtaining the identity letter of the user according to default identification algorithm Breath；

Prompt information sending module, if being configured as being determined the user for target use according to the identification information Family then sends study prompt information to the user；

Detection module is configured as detecting the target user for the response results of the study prompt information；

Display module, if being configured as detecting that the response results for confirmation study, are by system interaction interface The target user shows information relevant to the target object.

Optionally, the intelligent interaction device, further includes:

First Adapted information collection module is configured as collecting between the user and the target object according to preset condition Interactive instruction and Adapted information corresponding with the interactive instruction；

First model optimization module, be configured as by between the user and the target object interactive instruction and with it is described The corresponding Adapted information of interactive instruction is added to the training sample to form new training sample, based on the new training sample This carries out model optimization to the learning algorithm model.

Optionally, the intelligent interaction device, further includes:

Second Adapted information collection module is configured as collecting between the user and the target object according to preset condition Interactive instruction and the corresponding Adapted information of the interactive instruction and user to the adaptation corresponding with the interactive instruction The feedback score result of information；

Second model optimization module is configured as feedback score result being higher than the user and the institute of preset fraction threshold value It states the interactive instruction between target object and Adapted information corresponding with the interactive instruction is added to the training sample to be formed New training sample carries out model optimization to the learning algorithm model based on the new training sample.

Optionally, the interaction request receiving module, comprising:

Interaction request receiving submodule is configured as receiving the interaction request of user, carry in the interaction request State the voice question information that user is directed to the target object；

Speech recognition submodule is configured as carrying out speech recognition to the voice question information to obtain corresponding text Information；

The interactive instruction input module, comprising:

Information input submodule is configured as inputting the text information into learning algorithm model trained in advance.

Optionally, the interaction request receiving module, further includes:

Semantic analysis submodule is configured as carrying out semantic analysis to the text information, obtains corresponding semantic analysis As a result；

The information input submodule is also configured to inputting the semantic analysis result into study calculation trained in advance Method model.

Optionally, the interaction request receiving module, further includes:

Keyword extraction submodule is configured as carrying out keyword extraction to the text information according to preset rules；

The information input submodule is also configured to the keyword that will be extracted input study trained in advance and calculates Method model.

Optionally, the intelligent interaction device, further includes:

Data obtaining module is configured as:

If so, the information to acquisition is handled, and data store to treated.

Optionally, the data obtaining module, comprising:

Third model optimization submodule, is configured as that content is added to the training sample is new to be formed by treated Training sample carries out model optimization to the learning algorithm model based on the new training sample.

Optionally, the Adapted information sending module, further includes:

Language form analyzes submodule, is configured as analyzing the Adapted information, determines the Adapted information pair The language form answered；

Voice messaging converts submodule, is configured as passing through text based on preset speech samples library and the language form This transformation technology determines voice messaging corresponding with the Adapted information；

Voice messaging plays submodule, is configured as playing the voice messaging.

According to the another aspect of this specification embodiment, a kind of electronic equipment is provided, including memory, processor and deposit The computer instruction that can be run on a memory and on a processor is stored up, the processor realizes the intelligence when executing described instruction The step of energy exchange method.

According to the another aspect of this specification embodiment, a kind of computer readable storage medium is provided, is stored with meter The step of calculation machine instruction, which realizes the intelligent interactive method when being executed by processor.

In this specification embodiment, by the management for obtaining customer attribute information, target object attribute information, target object Scheme information, rudimentary knowledge information relevant to the target object and case data relevant with the target object；With Customer attribute information, target object attribute information, the administrative mechanism information of target object, basis relevant to the target object Knowledge information and case data relevant to the target object are training sample training learning algorithm model；Receive user's Interactive instruction between the user carried in the interaction request and the target object is inputted the study by interaction request Algorithm model obtains Adapted information relevant with the target object corresponding to the interactive instruction, and sends institute to user State Adapted information.

In this specification embodiment, pass through the learning algorithm mould that the interactive instruction input in interaction request is trained in advance Type exports corresponding Adapted information, is conducive to the high efficiency for improving interaction, while being also beneficial to improve the standard of the Adapted information Exactness.

Detailed description of the invention

Fig. 1 is the flow chart of intelligent interactive method provided by the embodiments of the present application；

Fig. 2 is the schematic diagram that link is called in model development deployment provided by the embodiments of the present application；

Fig. 3 is the schematic diagram that intelligent interactive method provided by the embodiments of the present application is applied to insurance scene；

Fig. 4 is the structural schematic diagram of intelligent interaction device provided by the embodiments of the present application；

Fig. 5 is the structural block diagram of electronic equipment provided by the embodiments of the present application.

Specific embodiment

Many details are explained in the following description in order to fully understand the application.But the application can be with Much it is different from other way described herein to implement, those skilled in the art can be without prejudice to the application intension the case where Under do similar popularization, therefore the application is not limited by following public specific implementation.

The term used in this specification one or more embodiment be only merely for for the purpose of describing particular embodiments, It is not intended to be limiting this specification one or more embodiment.In this specification one or more embodiment and appended claims The "an" of singular used in book, " described " and "the" are also intended to including most forms, unless context is clearly Indicate other meanings.It is also understood that term "and/or" used in this specification one or more embodiment refers to and includes One or more associated any or all of project listed may combine.

It will be appreciated that though may be retouched using term first, second etc. in this specification one or more embodiment Various information are stated, but these information should not necessarily be limited by these terms.These terms are only used to for same type of information being distinguished from each other It opens.For example, first can also be referred to as second, class in the case where not departing from this specification one or more scope of embodiments As, second can also be referred to as first.Depending on context, word as used in this " if " can be construed to " ... when " or " when ... " or " in response to determination ".

This specification embodiment provides a kind of intelligent interactive method.This specification is related to a kind of intelligent interaction dress simultaneously It sets, a kind of electronic equipment and a kind of computer readable storage medium are described in detail one by one in the following embodiments.

Fig. 1 shows the flow chart of the intelligent interactive method according to one embodiment of this specification, including step 102 is to step 108。

Step 102: receiving the interaction request of user, the user and the target object are carried in the interaction request Between interactive instruction.

In one embodiment that this specification provides, the interaction request includes project information counsel requests, project addition The interaction request of request or other forms；The project includes insurance coverage, mutual assistance project and public good project etc.；The interaction Instruction can be sent by button/option information of clicking of voice, text information or click robot interactive interface；If The user sends the interactive instruction by voice messaging, then after the interaction request of robot reception user, to institute's predicate Message breath carries out speech recognition to obtain corresponding text information；If the user can point by click robot interactive interface It hits button/option and sends the interactive instruction, then robot obtains corresponding text information by the button/option information.

In addition to this, after the interaction request for receiving user, also the user can be obtained according to default identification algorithm Identification information；Wherein, the identification algorithm is face recognition algorithm and/or algorithm for recognizing fingerprint.If according to institute It states identification information and determines that the user is target user, then send study prompt information to the user；Detect the mesh User is marked for the response results of the study prompt information；If detecting, the response results for confirmation study, pass through and are Interactive interface of uniting is that the target user shows information relevant to the target object.

By taking the project is insurance coverage as an example, it is assumed that the interaction request for receiving user is that " I wants to learn about health Danger ", after receiving the interaction request of user, robot can determine whether the user is M user by face recognition algorithm, If it is determined that the user is M user, and detect that there are contents to be learned in database, then sends study prompt letter to M user Breath；The M user is detected for the response results of the study prompt information；Specifically, M user can be handed over by clicking robot Mutual interface can push button choose whether to learn；If detecting, the response results for confirmation study, pass through robot Interactive interface is that the M user shows information relevant to the content to be learned.

Step 104: by the learning algorithm mould trained in advance of the interactive instruction input between the user and the target object Type.

In one embodiment that this specification provides, the learning algorithm model is trained in the following manner:

In one embodiment that this specification provides, model instruction is carried out based on artificial intelligence learning system (TensorFlow) To practice, it includes 6 links that the development deployment of the learning algorithm model, which calls link, it can specifically be realized by following steps:

1) sample prepares

The operation of general TensorFlow the application code definition comprising chart (Graph) and session control (Session), Size of code is little, can be encapsulated into a file.It needs to prepare sample data and test data, GDF general data file before training It is space or comma-delimited file (CSV).

2) Feature Engineering

Feature Engineering refers to filters out better data characteristics with a series of mode of engineering from initial data, to mention The training effect of rising mould type.Feature Engineering generally includes the links such as data prediction, feature selecting, dimensionality reduction.

3) model training

Model training is to adjust model parameter by training data, the mistake for improving model for the fitting degree of data Journey.

4) model is disposed

Artificial intelligence learning service system (TensorFlow Serving) is one for machine learning model service (serving) high-performance open source library.Trained machine learning model can be deployed on line by it, use remote process tune System (gRPC) is used as interface and receives external call.It is updated and automodel version management in addition, it goes back support model heat.Tool Body, after completing model training, model deployment need to only can be realized by simple program.

5) model calls

After having trained a model, in order to reuse later, usually we need to save the result of model.Such as Fruit is gone to realize neural network with Tensorflow, and what is saved is exactly every weighted value in neural network.Later period is using When model, can directly it be called by model of the code to preservation.

6) log flows back

Log reflux by feature log recording and flows back.

This specification provide one embodiment in, model development deployment call link by sample preparation, Feature Engineering, This 6 links that model training, model are disposed, model calls, log flows back, form closed loop, as shown in Figure 2.Form the meaning of closed loop Justice is to record marking feature online, convenient for the generation of sample and characteristic, while guaranteeing the strong consistency of data.

It, will be with customer attribute information, target object attribute information, target object during learning algorithm model training Administrative mechanism information, rudimentary knowledge information relevant to the target object and case number of cases relevant with the target object It is trained according to the learning algorithm model is input to as training sample.

In one embodiment that this specification provides, the intelligent interactive method can be applied to the user of any scene with The interaction of intelligent robot, by taking the intelligent interactive method is applied to insurance consulting scene as an example, the target object is to protect Danger, the target object attribute information is Insurance Attribute information, and the administrative mechanism of target object is insurance management mechanism, with The relevant rudimentary knowledge information of target object and case data relevant to the target object are respectively and insurance phase The rudimentary knowledge information of pass and case data relevant to insurance.

The training method of learning algorithm model through this embodiment, with customer attribute information, target object attribute information, The administrative mechanism information of target object, rudimentary knowledge information relevant to the target object and related with the target object Case data be that training sample trains learning algorithm model, thus realize customer attribute information, target object attribute information, The administrative mechanism information of target object, rudimentary knowledge information relevant to the target object and related with the target object Case data between association.Also, due to by with customer attribute information, target object attribute information, target object Administrative mechanism information, rudimentary knowledge information relevant to the target object and case data relevant with the target object Training pattern, so that can more embody the Adapted information of the user and model output in the use process of learning algorithm model Between the degree of association.

In one embodiment that this specification provides, the interactive instruction between the user and the target object can pass through Button/option information of clicking of voice, text information or the click robot interactive interface is sent.Assuming that user is logical The interactive instruction that voice sends " condition of insuring of health insurance " to intelligent robot is crossed, the intelligent robot receives the friendship of user Mutually after instruction, speech recognition is carried out to obtain corresponding text information to get the text arrived to the voice messaging of the user Information is " condition of insuring of health insurance ", and the text information is then inputted to learning algorithm trained in advance by intelligent robot Model.

Specifically, by before text information input learning algorithm model trained in advance, it can also be to the text envelope Breath carries out semantic analysis, obtains corresponding semantic analysis result, and the study that semantic analysis result input is trained in advance Algorithm model.

Still by taking the intelligent interactive method is applied to insurance consulting scene as an example, it is assumed that user passes through voice to intelligence machine After human hair send the interactive instruction of " condition of insuring of health insurance ", the intelligent robot to receive the interactive instruction of user, to institute It is " health insurance that the voice messaging for stating user, which carries out speech recognition to obtain corresponding text information to get the text information arrived, Insure condition ", semantic analysis then is carried out to the text information by intelligent robot, obtaining corresponding semantic analysis result is " condition of insuring of health insurance ", and the learning algorithm model that semantic analysis result input is trained in advance.

In addition to this, by before text information input learning algorithm model trained in advance, acceptable basis is default Rule carries out keyword extraction, and the learning algorithm mould that the input of the keyword of extraction is trained in advance to the text information Type.

It uses the example above, it is assumed that user is referred to by voice to the interaction that intelligent robot sends " condition of insuring of health insurance " It enables, after the intelligent robot receives the interactive instruction of user, speech recognition is carried out to obtain to the voice messaging of the user To corresponding text information to get to text information be " condition of insuring of health insurance ", then by intelligent robot to described Text information carries out keyword extraction, it is assumed that the keyword of extraction is " health insurance, condition of insuring ", by the keyword of extraction Input learning algorithm model trained in advance.

Step 106: and the target object phase corresponding with the interactive instruction is exported by the learning algorithm model The Adapted information of pass.

In one embodiment that this specification provides, Adapted information is answer information relevant to the interaction request of user, By with the administrative mechanism information, related to the target object of customer attribute information, target object attribute information, target object Rudimentary knowledge information and relevant to target object case data be that training sample trains learning algorithm model, from And realize the administrative mechanism information, related to the target object of customer attribute information, target object attribute information, target object Rudimentary knowledge information and case data relevant to the target object between association.

Still by taking the intelligent interactive method is applied to insurance consulting scene as an example, it is assumed that user passes through voice to intelligence machine Human hair send the interactive instruction of " I wants to learn about health insurance ", and the intelligent robot receives the interactive instruction of user, to described The voice messaging of user carries out speech recognition to obtain corresponding text information, then carries out semantic point to the text information Analysis, obtaining corresponding semantic analysis result is " health insurance ", and semantic analysis result input study trained in advance is calculated Method model relevant with the target object is adapted to letter by the way that learning algorithm model output is corresponding to the interactive instruction Breath.

Assuming that the learning algorithm model is handled the result exported according to semantic analysis result " health insurance " is " health insurance is the Chinese abbreviation of health insurance, refers to that insurance company is protected by sickness insurance, medical insurance, disability revenue losses Insurance of the modes such as danger and nursing insurance to loss payment insurance money caused by through poor health ".

It should be noted that described above is the preferred embodiment of technical scheme, some of them step is simultaneously It is not necessary to realizing technical scheme.Model once establishes, whithin a period of time can Reusability on line, Adapted information in order to guarantee model output is more acurrate, and the related data that can periodically choose update re-establishes model, but simultaneously It is not to provide the required step of Adapted information every time for the user.

Step 108: Xiang Suoshu user sends the Adapted information.

In one embodiment that this specification provides, the Adapted information of the learning algorithm model output is written form, The Adapted information can be sent to the user by way of text, can also by way of voice, picture or table into Row is sent.

Assuming that the Adapted information is sent by way of voice, then specific information conversion process can pass through following steps It realizes:

Based on the speech samples library and the language form prestored, generated and the Adapted information by text transformation technology Corresponding voice messaging.

After converting voice messaging for Adapted information, Xiang Suoshu user sends the Adapted information and passes through intelligent robot Play the voice messaging.

Still by taking the intelligent interactive method is applied to insurance consulting scene as an example, it is assumed that robot receives the interaction of user Request is voice messaging " condition of insuring of insurance ", after receiving the interaction request of user, to the voice messaging of the user Speech recognition is carried out to obtain corresponding text information, keyword extraction then is carried out to the text information, it is assumed that extraction Keyword is " condition of insuring ", by the keyword input of extraction learning algorithm model trained in advance, it is assumed that the study Algorithm model is that " condition of insuring " is handled the result exported as " insurer must have corresponding power according to keyword Sharp ability and capacity, otherwise ordered insurance contract does not become legally effective ", i.e., it is defeated by the learning algorithm model Adapted information relevant with the target object corresponding with the interactive instruction is that " insurer must have corresponding right out Ability and capacity, otherwise ordered insurance contract does not become legally effective ".

After determining Adapted information, to the Adapted information " insurer must have corresponding legal capacity and capacity, Otherwise ordered insurance contract does not become legally effective " it is analyzed, determine that the corresponding language form of the Adapted information is Chinese is generated and the Adapted information then based on the speech samples library and the language form prestored by text transformation technology Corresponding voice messaging, and " insurer must have corresponding right by the intelligent robot voice broadcasting Adapted information Ability and capacity, otherwise ordered insurance contract does not become legally effective ".

In one embodiment that this specification provides, after the completion of the learning model training, in order to guarantee model output The accuracy of Adapted information, the related data that can periodically choose update carry out model optimization, and specific model optimization process can It is realized by following steps:

Specifically, the preset condition is the preset period, it is assumed that predetermined period is 10 days, collects institute according to preset condition It states the interactive instruction between user and the target object and Adapted information corresponding with the interactive instruction collects institute in 10 days State the interactive instruction between user and the target object and Adapted information corresponding with the interactive instruction.After the completion of information is collected The information of collection is added to the training sample to form new training sample, based on the new training sample to It practises algorithm model and carries out model optimization.

Preset condition described in practical application can also be that the quantity of the Adapted information, the present invention are without limitation.

In addition to this, the model optimization process can also be realized by following steps:

Specifically, assuming that the interactive instruction of the user is " condition of insuring of insurance ", corresponding with the interactive instruction Adapted information is that " insurer must have corresponding legal capacity and capacity, and otherwise ordered insurance contract does not occur Legal effect ", user are 8 points to the feedback score of the Adapted information；Assuming that the interactive instruction of the user is that " I wants to understand Health insurance once ", Adapted information corresponding with the interactive instruction are that " health insurance is the Chinese abbreviation of health insurance, refers to guarantor Dangerous company leads to through poor health by sickness insurance, medical insurance, the disability modes such as loss of income insurance and nursing insurance Loss payment insurance money insurance ", user is 9 points to the feedback score of the Adapted information.It, will be anti-after the completion of information is collected Feedback appraisal result is higher than the interactive instruction between the user and the target object of preset fraction threshold value and refers to the interaction Corresponding Adapted information is enabled to be added to the training sample to form new training sample, based on the new training sample to institute It states learning algorithm model and carries out model optimization.

In one embodiment that this specification provides, due to rudimentary knowledge relevant to the target object and the mesh Mark object administrative mechanism information constantly updating, therefore, in order to guarantee model output Adapted information accuracy, Ke Yiding Phase obtains the information updated and carries out model optimization, and specific information access process can be realized by following steps:

If so, the information to acquisition is handled, and data store to treated.

Specifically, the information to acquisition is handled, and content is added to the training sample to be formed by treated New training sample carries out model optimization to the learning algorithm model based on the new training sample.

It is that " obtaining clicking rate in 24 hours is more than 10,000 with the preset rules in one embodiment that this specification provides For the information for including in secondary webpage ", it is assumed that the target object is english article, and the webpage for meeting the preset rules is Webpage B, webpage C and webpage D, then obtain the URL of webpage B, webpage C and webpage D, by web page analysis algorithm the webpage B, It is extracted in webpage C and webpage D and relevant to english article link and described link is added to obtaining for URL queue outstanding message It takes；It, can be true if the english article quantity of the acquisition reaches preset 100 after obtaining a certain number of english articles Surely reach preset acquisition information task and stop execution condition, then stop obtaining english article, and to the english article got It is handled and is stored.

In one embodiment that this specification provides, after the interaction request for receiving user, identified according to identification algorithm The identity information of user provides learning functionality for target user, be conducive to help target user learn in time it is newest with it is described The relevant rudimentary knowledge information of target object and administrative mechanism information, and learning cost is effectively reduced, improve study Efficiency；Meanwhile after the completion of model training, model optimization is carried out according to the relevant information of continuous renewal, advantageously ensures that model The accuracy of output information；The program or script of web message, automatic collection institute are automatically grabbed according to certain rules There is the content of pages being able to access that, to obtain or update the content and retrieval mode of these websites, advantageously ensures that database In relevant to target object information can be timely updated.

Fig. 3 shows the intelligent interactive method of one embodiment of this specification, and the intelligent interactive method to be to insurance scene Example is described, including step 302 is to step 318.

Step 302: receiving the interaction request of user, official communication of the user to insurance coverage is carried in the interaction request Ask information.

Step 304: the identification information of the user is obtained according to face recognition algorithm.

Step 306: judging whether the user is insurance agent, if so, thening follow the steps 308；If it is not, then executing step Rapid 314.

Step 308: Xiang Suoshu user sends study prompt information.

Step 310: whether detection response results are confirmation study, if so, thening follow the steps 312；If it is not, thening follow the steps 314。

Specifically, whether detect the target user is confirmation study for the response results of the study prompt information, If so, thening follow the steps 312；If it is not, thening follow the steps 314.

Step 312: being that the target user shows information relevant to insurance by system interaction interface.

Specifically, the interaction request for assuming to receive user is " I wants to learn about health insurance ", receiving user's After interaction request, robot can determine whether the user is party A-subscriber by face recognition algorithm, however, it is determined that the user is A User, and detect that there are contents to be learned in database, then study prompt information is sent to party A-subscriber；Detect the party A-subscriber couple In the response results of the study prompt information；Specifically, party A-subscriber can by click robot interactive interface can push button To choose whether to learn；If detecting, the response results are the party A-subscriber by robot interactive interface for confirmation study Show information relevant to the content to be learned.

Specifically, the interaction request can by voice, text information or click the robot interactive interface can Button/option information is clicked to be sent；If the user sends the interaction request by voice, robot receives user Interaction request after, to the voice messaging carry out speech recognition to obtain corresponding text information；If the user passes through Button/the option of clicking at robot interactive interface sends the interactive instruction, then robot passes through the button/option information Obtain corresponding text information.

Step 314: by consultation information input of the user to insurance coverage learning algorithm model trained in advance.

Obtain customer attribute information, Insurance Attribute information, the administrative mechanism information of insurance, rudimentary knowledge relevant to insurance Information and case data relevant to insurance；

With customer attribute information, Insurance Attribute information, the administrative mechanism information of insurance, rudimentary knowledge relevant to insurance letter Breath and case data relevant to insurance are training sample training learning algorithm model.

The training method of learning algorithm model through this embodiment, with customer attribute information, Insurance Attribute information, insurance Administrative mechanism information, rudimentary knowledge information relevant to insurance and to be training sample instruct with relevant case data is insured Practice learning algorithm model, thus realize customer attribute information, Insurance Attribute information, the administrative mechanism information of insurance, with insure phase Association between the rudimentary knowledge information of pass and case data relevant to insurance.Also, due to by being believed with user property Breath, Insurance Attribute information, the administrative mechanism information of insurance, rudimentary knowledge information relevant to insurance and the relevant case with insurance Number of cases is fitted so that can more embody the user in the use process of learning algorithm model with the described of model output according to training pattern With the degree of association between information.

This specification provide one embodiment in, the user to the consultation information of insurance coverage can by voice, Text information or the button/option information of clicking for clicking the robot interactive interface are sent.Assuming that user passes through language Sound sends the interaction request of " I wants to learn about health insurance " to intelligent robot, and the intelligent robot receives the interaction of user After request, speech recognition is carried out to obtain corresponding text information to get the text envelope arrived to the voice messaging of the user Breath is " I wants to learn about health insurance ", and the text information is then inputted to learning algorithm trained in advance by intelligent robot Model.

Assuming that it is that " I thinks that intelligent robot, which carries out the text information that speech recognition obtains to the voice messaging of the user, Solution once health insurance " then carries out keyword extraction to the text information by intelligent robot, it is assumed that the keyword of extraction is " health insurance ", by the keyword input of extraction learning algorithm model trained in advance.

Step 316: Adapted information corresponding with the interaction request is exported by the learning algorithm model.

It uses the example above, the intelligent robot receives the interaction request of user, carries out language to the voice messaging of the user Sound is identified to obtain corresponding text information, then carries out keyword extraction to the text information, obtains corresponding extraction knot Fruit is " health insurance ", and the learning algorithm model that the input of the keyword of extraction is trained in advance, it is assumed that the learning algorithm It is " basic information of health insurance " that model, which is handled the result exported according to keyword " health insurance ",.

Step 318: Xiang Suoshu user sends the Adapted information.

Assuming that the Adapted information is sent by way of voice, then after converting voice messaging for Adapted information, to institute It states user and sends the Adapted information and pass through intelligent robot and play the voice messaging, is i.e. intelligent robot is by health insurance Rudimentary knowledge passes to user by way of voice plays.

In one embodiment that this specification provides, after the completion of the learning model training, in order to guarantee model output The accuracy of Adapted information, the related data that can periodically choose update carry out model optimization, and specific model optimization process can Referring to implementation described in Fig. 1 method, details are not described herein.

In one embodiment that this specification provides, after the interaction request for receiving user, identified according to face recognition algorithm The identity information of user provides learning functionality for target user, is conducive to that target user is helped to learn newest and insurance in time Relevant rudimentary knowledge information and administrative mechanism information, and learning cost is effectively reduced, improve learning efficiency；Together When, after the completion of model training, model optimization is carried out according to the relevant information of continuous renewal, advantageously ensures that model output information Accuracy；By constantly obtaining new information automatically, the energy of information relevant to target object in database is advantageously ensured that It accesses and timely updates.

Corresponding with above method embodiment, this specification additionally provides intelligent interaction device embodiment, and Fig. 4 shows this The structural schematic diagram of the intelligent interaction device of specification one embodiment.As shown in figure 4, the device includes:

Interaction request receiving module 402 is configured as receiving the interaction request of user, carry in the interaction request State the interactive instruction between user and the target object；

Interactive instruction input module 404 is configured as the interactive instruction input between the user and the target object Trained learning algorithm model in advance；

Adapted information output module 406 is configured as exporting and the interactive instruction pair by the learning algorithm model The Adapted information relevant to the target object answered；

Adapted information sending module 408 is configured as sending the Adapted information to the user.

Optionally, the intelligent interaction device, further includes:

Optionally, the interaction request receiving module, comprising:

The interactive instruction input module, comprising:

Optionally, the interaction request receiving module, further includes:

Optionally, the intelligent interaction device, further includes:

Data obtaining module is configured as:

The uniform resource location (URL) of Initial page is obtained according to preset rules；

Link and by institute relevant to the target object is extracted in the Initial page by web page analysis algorithm State the acquisition for linking and being added to URL queue outstanding message relevant to the target object；

Judge whether that reaching the default information task that obtains stops execution condition, if it is not, then repeating according to preset rules The step of obtaining uniform resource location (URL) of Initial page；

If so, the information to acquisition is handled, and data store to treated.

Optionally, the data obtaining module, comprising:

Optionally, the Adapted information sending module, further includes:

Voice messaging converts submodule, is configured as passing through text based on preset speech samples library and the language form This transformation technology (TTS) determines voice messaging corresponding with the Adapted information；

Voice messaging plays submodule, is configured as playing the voice messaging.

Fig. 5 shows the structural block diagram of the electronic equipment 500 according to one embodiment of this specification.The electronic equipment 500 Component includes but is not limited to memory 510 and processor 520.Processor 520 is connected with memory 510 by bus 530, number According to library 550 for saving data.

Electronic equipment 500 further includes access device 540, access device 540 enable electronic equipment 500 via one or Multiple networks 560 communicate.The example of these networks includes public switched telephone network (PSTN), local area network (LAN), wide area network (WAN), the combination of the communication network of personal area network (PAN) or such as internet.Access device 540 may include wired or wireless One or more of any kind of network interface (for example, network interface card (NIC)), such as IEEE802.11 wireless local area Net (WLAN) wireless interface, worldwide interoperability for microwave accesses (Wi-MAX) interface, Ethernet interface, universal serial bus (USB) connect Mouth, cellular network interface, blue tooth interface, near-field communication (NFC) interface, etc..

In one embodiment of this specification, other unshowned portions in the above-mentioned component and Fig. 5 of electronic equipment 500 Part can also be connected to each other, such as pass through bus.It should be appreciated that electronic devices structure block diagram shown in fig. 5 merely for the sake of Exemplary purpose, rather than the limitation to this specification range.Those skilled in the art can according to need, and increases or replaces it His component.

Electronic equipment 500 can be any kind of static or mobile electronic device, including mobile computer or mobile electricity Sub- equipment (for example, tablet computer, personal digital assistant, laptop computer, notebook computer, net book etc.), movement Phone (for example, smart phone), wearable electronic equipment (for example, smartwatch, intelligent glasses etc.) or other kinds of shifting Dynamic equipment, or the stationary electronic devices of such as desktop computer or PC.Electronic equipment 500 can also be mobile or state type Server.

Wherein, processor 520 is for executing the step of intelligent interactive method as previously described can be performed in following computer.

The exemplary scheme of the above-mentioned a kind of electronic equipment for the present embodiment.It should be noted that the skill of the electronic equipment Art scheme and the technical solution of above-mentioned intelligent interactive method belong to same design, and the technical solution of electronic equipment is not described in detail Detail content, may refer to the description of the technical solution of above-mentioned intelligent interactive method.

One embodiment of this specification also provides a kind of computer readable storage medium, is stored with computer instruction, this refers to The step of intelligent interactive method as previously described is realized when order is executed by processor.

A kind of exemplary scheme of above-mentioned computer readable storage medium for the present embodiment.It should be noted that this is deposited The technical solution of storage media and the technical solution of above-mentioned intelligent interactive method belong to same design, the technical solution of storage medium The detail content being not described in detail may refer to the description of the technical solution of above-mentioned intelligent interactive method.

It is above-mentioned that this specification specific embodiment is described.Other embodiments are in the scope of the appended claims It is interior.In some cases, the movement recorded in detail in the claims or step can be come according to the sequence being different from embodiment It executes and desired result still may be implemented.In addition, process depicted in the drawing not necessarily require show it is specific suitable Sequence or consecutive order are just able to achieve desired result.In some embodiments, multitasking and parallel processing be also can With or may be advantageous.

The computer instruction includes computer program code, the computer program code can for source code form, Object identification code form, executable file or certain intermediate forms etc..The computer-readable medium may include: that can carry institute State any entity or device, recording medium, USB flash disk, mobile hard disk, magnetic disk, CD, the computer storage of computer program code Device, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), Electric carrier signal, telecommunication signal and software distribution medium etc..It should be noted that the computer-readable medium include it is interior Increase and decrease appropriate can be carried out according to the requirement made laws in jurisdiction with patent practice by holding, such as in certain jurisdictions of courts Area does not include electric carrier signal and telecommunication signal according to legislation and patent practice, computer-readable medium.

It should be noted that for the various method embodiments described above, describing for simplicity, therefore, it is stated as a series of Combination of actions, but those skilled in the art should understand that, the application is not limited by the described action sequence because According to the application, certain steps can use other sequences or carry out simultaneously.Secondly, those skilled in the art should also know It knows, the embodiments described in the specification are all preferred embodiments, and related actions and modules might not all be this Shen It please be necessary.

In the above-described embodiments, it all emphasizes particularly on different fields to the description of each embodiment, there is no the portion being described in detail in some embodiment Point, it may refer to the associated description of other embodiments.

The application preferred embodiment disclosed above is only intended to help to illustrate the application.There is no detailed for alternative embodiment All details are described, are not limited the invention to the specific embodiments described.Obviously, according to the content of this specification, It can make many modifications and variations.These embodiments are chosen and specifically described to this specification, is in order to preferably explain the application Principle and practical application, so that skilled artisan be enable to better understand and utilize the application.The application is only It is limited by claims and its full scope and equivalent.

Claims

1. a kind of intelligent interactive method characterized by comprising

The interaction request of user is received, the interaction carried between the user and the target object in the interaction request refers to It enables；

Adapted information relevant with the target object corresponding to the interactive instruction is exported by the learning algorithm model；

The Adapted information is sent to the user.

2. the method according to claim 1, wherein the learning algorithm model is instructed in the following manner Practice:

Obtain customer attribute information, target object attribute information, target object administrative mechanism information, with the target object phase The rudimentary knowledge information of pass and case data relevant to the target object；

With the administrative mechanism information, related to the target object of customer attribute information, target object attribute information, target object Rudimentary knowledge information and case data relevant to the target object be training sample training learning algorithm model.

3. the method according to claim 1, wherein it is described receive user interaction request step execute after, Further include:

If determining that the user is target user according to the identification information, study prompt letter is sent to the user Breath；

If detect the response results for confirmation study, by system interaction interface be the target user show with it is described The relevant information of target object.

4. the method according to claim 1, wherein described hold to the user transmission Adapted information step After row, further includes:

Interactive instruction between the user and the target object and corresponding with the interactive instruction is collected according to preset condition Adapted information；

By between the user and the target object interactive instruction and Adapted information corresponding with the interactive instruction be added to The training sample carries out model to the learning algorithm model to form new training sample, based on the new training sample Optimization.

5. the method according to claim 1, wherein described hold to the user transmission Adapted information step After row, further includes:

Interactive instruction between the user and the target object, corresponding with the interactive instruction suitable is collected according to preset condition With information and user to the feedback score result of the Adapted information corresponding with the interactive instruction；

By feedback score result be higher than preset fraction threshold value the user and the target object between interactive instruction and with institute It states the corresponding Adapted information of interactive instruction and is added to the training sample to form new training sample, based on the new training Sample carries out model optimization to the learning algorithm model.

6. the method according to claim 1, wherein the interactive instruction is voice question information；

The interactive instruction by between the user and the target object inputs learning algorithm model trained in advance

By text information input learning algorithm model trained in advance.

7. according to the method described in claim 6, it is characterized in that, the study that text information input is trained in advance Before algorithm model sub-step executes, further includes:

By semantic analysis result input learning algorithm model trained in advance.

8. according to the method described in claim 6, it is characterized in that, the study that text information input is trained in advance Before algorithm model sub-step executes, further includes:

By the keyword input of extraction learning algorithm model trained in advance.

9. the method according to claim 1, wherein further include:

Extracted in the Initial page by web page analysis algorithm it is relevant to the target object link and will it is described with The target object is relevant to link the acquisition for being added to uniform resource location queue outstanding message；

Judge whether that reaching the default information task that obtains stops execution condition, obtains if it is not, then repeating according to preset rules The step of uniform resource location of Initial page；

If so, the information to acquisition is handled, and data store to treated.

10. according to the method described in claim 9, it is characterized in that, the information of described pair of acquisition carries out processing sub-step execution Afterwards, further includes:

By treated, content is added to the training sample to form new training sample, based on the new training sample pair The learning algorithm model carries out model optimization.

11. the method according to claim 1, wherein it is described by the learning algorithm model output with it is described After the corresponding Adapted information step relevant to the target object of interactive instruction executes, Xiang Suoshu user sends the adaptation Before information Step executes, further includes:

Based on the speech samples library and the language form prestored, generated by text transformation technology corresponding with the Adapted information Voice messaging；

It is described to include: to the user transmission Adapted information

Play the voice messaging.

12. a kind of intelligent interaction device characterized by comprising

Interaction request receiving module is configured as receiving the interaction request of user, carries the user in the interaction request With the interactive instruction between the target object；

Interactive instruction input module is configured as the interactive instruction input training in advance between the user and the target object Learning algorithm model；

Adapted information output module is configured as exporting and institute corresponding with the interactive instruction by the learning algorithm model State the relevant Adapted information of target object；

13. device according to claim 12, which is characterized in that further include:

Module is obtained, is configured as obtaining the administrative mechanism letter of customer attribute information, target object attribute information, target object Breath, rudimentary knowledge information relevant to the target object and case data relevant with the target object；

Model training module is configured as believing with the administrative mechanism of customer attribute information, target object attribute information, target object Breath, rudimentary knowledge information relevant to the target object and case data relevant with the target object are training sample Training learning algorithm model.

14. device according to claim 12, which is characterized in that further include:

Identity information acquisition module is configured as obtaining the identification information of the user according to default identification algorithm；

Prompt information sending module, if being configured as determining that the user is target user according to the identification information, Study prompt information is sent to the user；

Display module, if being configured as detecting that the response results are described by system interaction interface for confirmation study Target user shows information relevant to the target object.

15. device according to claim 12, which is characterized in that further include:

First Adapted information collection module is configured as collecting the friendship between the user and the target object according to preset condition Mutually instruction and Adapted information corresponding with the interactive instruction；

First model optimization module, be configured as by between the user and the target object interactive instruction and with the interaction Corresponding Adapted information is instructed to be added to the training sample to form new training sample, based on the new training sample pair The learning algorithm model carries out model optimization.

16. device according to claim 12, which is characterized in that further include:

Second Adapted information collection module is configured as collecting the friendship between the user and the target object according to preset condition Mutually instruction and the corresponding Adapted information of the interactive instruction and user are to the Adapted information corresponding with the interactive instruction Feedback score result；

Second model optimization module is configured as feedback score result being higher than the user of preset fraction threshold value and the mesh It is new to be formed that interactive instruction and Adapted information corresponding with the interactive instruction between mark object are added to the training sample Training sample carries out model optimization to the learning algorithm model based on the new training sample.

17. device according to claim 12, which is characterized in that the interaction request receiving module, comprising:

Interaction request receiving submodule is configured as receiving the interaction request of user, carries the use in the interaction request Family is directed to the voice question information of the target object；

Speech recognition submodule is configured as carrying out speech recognition to the voice question information to obtain corresponding text envelope Breath；

The interactive instruction input module, comprising:

18. device according to claim 17, which is characterized in that the interaction request receiving module, further includes:

Semantic analysis submodule is configured as carrying out semantic analysis to the text information, obtains corresponding semantic analysis result；

The information input submodule is also configured to inputting the semantic analysis result into learning algorithm mould trained in advance Type.

19. device according to claim 17, which is characterized in that the interaction request receiving module, further includes:

The information input submodule is also configured to the keyword that will be extracted input learning algorithm mould trained in advance Type.

20. device according to claim 12, which is characterized in that further include:

Data obtaining module is configured as:

If so, the information to acquisition is handled, and data store to treated.

21. device according to claim 20, which is characterized in that the data obtaining module, comprising:

Third model optimization submodule, is configured as that content is added to the training sample to form new training by treated Sample carries out model optimization to the learning algorithm model based on the new training sample.

22. device according to claim 12, which is characterized in that the Adapted information sending module, further includes:

Language form analyzes submodule, is configured as analyzing the Adapted information, determines that the Adapted information is corresponding Language form；

Voice messaging converts submodule, is configured as turning based on preset speech samples library and the language form by text Change technology determines voice messaging corresponding with the Adapted information；

Voice messaging plays submodule, is configured as playing the voice messaging.

23. a kind of electronic equipment characterized by comprising

Memory, processor；

The memory is for storing computer executable instructions, and for executing, the computer is executable to be referred to the processor It enables:

The Adapted information is sent to the user.

24. a kind of computer readable storage medium, is stored with computer instruction, which is characterized in that the instruction is held by processor The step of claim 1-11 any one the method is realized when row.