CN109582773A - Intelligent answer matching process and device - Google Patents
Intelligent answer matching process and device Download PDFInfo
- Publication number
- CN109582773A CN109582773A CN201811448785.3A CN201811448785A CN109582773A CN 109582773 A CN109582773 A CN 109582773A CN 201811448785 A CN201811448785 A CN 201811448785A CN 109582773 A CN109582773 A CN 109582773A
- Authority
- CN
- China
- Prior art keywords
- subject information
- value
- history
- history subject
- real
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The present embodiments relate to technical field of information processing, in particular to a kind of intelligent answer matching process and device.This method can realize multi-angular analysis and the processing to question information by convolutional neural networks model, machine learning model, and then filter out qualified history subject information, and calculate the score value of these history subject informations, dual judgement is carried out based on score value to obtain the question and answer pair for meeting decision condition, it so, it is possible the quality for the question and answer pair that raising acquires.
Description
Technical field
The present embodiments relate to technical field of information processing, in particular to a kind of intelligent answer matching process and
Device.
Background technique
The earliest realization conception of question answering system can trace back to turing test.In order to which whether test machine has mankind's intelligence
Can, turing test requires computer that can answer a series of problems proposed by mankind tester in 5 minutes, and it is reached more than
30% answer allows tester to be mistakenly considered the mankind and is answered.With the development of the relevant technologies such as artificial intelligence, natural language processing,
Variation for different data shapes also derives different types of question answering system.But existing question answering system is directed to user
The new problem processing mode of proposition is single, and flexibility is not high, and then causes the answer matched of low quality.
Summary of the invention
In view of this, the present invention provides a kind of intelligent answer matching process and devices.
The embodiment of the invention provides a kind of intelligent answer matching process, applied to the service connecting with client communication
End, which comprises
The question information that the client is sent is received, the real-time subject information of the question information is extracted；
Convolutional neural networks model is constructed, all history subject informations in presetting database are refreshing to the convolution is inputted
It is trained through network model；
The multiple history subject informations to match with the real-time subject information are found out from the presetting database,
In, the first similarity between each history subject information and the real-time subject information in the multiple history subject information
Value is less than first threshold；The multiple history subject information is inputted into the convolutional neural networks model of training completion to calculate
State the second similarity value between each history subject information and the real-time subject information in multiple history subject informations；Base
In multiple second similarity values being calculated, preset quantity is chosen from the multiple history subject information according to setting sequence
A history subject information；
According to the building of the multiple history subject information and training machine learning model, by the real-time subject information and institute
It states the preset quantity history subject information input machine learning model and the preset quantity history theme letter is calculated
The score value of each history subject information in breath；
Judge whether the maximum value in the multiple score values being calculated reaches setting value, if the maximum value reaches described
Setting value calculates the editing distance value between the corresponding history subject information of the maximum value and the real-time subject information；
Judge whether the editing distance value is less than second threshold, if the editing distance value is less than the second threshold,
The question and answer pair for obtaining the corresponding history subject information of the maximum value, by the question and answer to being sent to the client.
Optionally, the method also includes:
If the maximum value does not reach the setting value, searches whether to exist and match with the real-time subject information
The corresponding client of user's portrait；If it exists, the question information is sent to the client found out so that the visitor found out
Family end is answered by the question information.
Optionally, based on multiple second similarity values being calculated, according to setting sequence from the multiple history theme
The step of preset quantity history subject information is chosen in information, comprising:
Multiple second similarity values being calculated are ranked up according to sequence from high to low；
Obtain the corresponding history subject information of the second similarity value of the forward setting quantity that sorts.
Optionally, according to the building of the multiple history subject information and training machine learning model the step of, comprising:
For each history subject information in the multiple history subject information, obtains in the history subject information and include
Label characteristics value, characteristic of division value and similarity characteristic value；
Building machine learning model, the multiple labels that will acquire are established based on gradient boosted tree and algorithm with regress analysis method
Characteristic value, multiple characteristic of division values and multiple similarity characteristic values input the machine learning model and are trained.
Optionally, the real-time subject information and the preset quantity history subject information are inputted into the machine learning
The step of score value of each history subject information in the preset quantity history subject information is calculated in model, packet
It includes:
Obtain the label characteristics value, characteristic of division value and similarity characteristic value of the real-time subject information；
Label characteristics value, the classification for obtaining each history subject information in the preset quantity history subject information are special
Value indicative and similarity characteristic value；
By the label characteristics value of the real-time subject information, characteristic of division value and similarity characteristic value and the present count
Measure the label characteristics value, characteristic of division value and the input of similarity characteristic value of each history subject information in a history subject information
Each history subject information in the preset quantity history subject information is calculated in the machine learning model that training is completed
Score value.
The embodiment of the invention also provides a kind of intelligent answer coalignments, applied to the service connecting with client communication
End, described device include:
Real-time subject information extraction module, the question information sent for receiving the client extract the enquirement letter
The real-time subject information of breath；
Convolutional neural networks model construction module, for constructing convolutional neural networks model, by the institute in presetting database
There is history subject information to be trained to the convolutional neural networks model is inputted；
Subject information screening module matches for finding out from the presetting database with the real-time subject information
Multiple history subject informations, wherein each history subject information in the multiple history subject information and the real-time master
The first similarity value inscribed between information is less than first threshold；The multiple history subject information is inputted into the convolution that training is completed
Neural network model is believed with calculating each history subject information in the multiple history subject information and the real-time theme
The second similarity value between breath；Based on multiple second similarity values being calculated, gone through according to setting sequence from the multiple
Preset quantity history subject information is chosen in history subject information；
Score value computing module will for constructing simultaneously training machine learning model according to the multiple history subject information
The real-time subject information and the preset quantity history subject information input machine learning model are calculated described
The score value of each history subject information in preset quantity history subject information；
First judgment module, for judging whether the maximum value in the multiple score values being calculated reaches setting value, if
The maximum value reaches the setting value, calculate the corresponding history subject information of the maximum value and the real-time subject information it
Between editing distance value；
Second judgment module, for judging whether the editing distance value is less than second threshold, if the editing distance value
Less than the second threshold, the question and answer pair of the corresponding history subject information of the maximum value are obtained, by the question and answer to being sent to
The client.
Optionally, the first judgment module is also used to:
If the maximum value does not reach the setting value, searches whether to exist and match with the real-time subject information
The corresponding client of user's portrait；If it exists, the question information is sent to the client found out so that the visitor found out
Family end is answered by the question information.
Optionally, the subject information screening module is in the following manner based on multiple second similarities being calculated
Value, chooses preset quantity history subject information according to setting sequence from the multiple history subject information:
Multiple second similarity values being calculated are ranked up according to sequence from high to low；
Obtain the corresponding history subject information of the second similarity value of the forward setting quantity that sorts.
Optionally, the score value computing module is constructed and is instructed according to the multiple history subject information in the following manner
Practice machine learning model:
For each history subject information in the multiple history subject information, obtains in the history subject information and include
Label characteristics value, characteristic of division value and similarity characteristic value；
Building machine learning model, the multiple labels that will acquire are established based on gradient boosted tree and algorithm with regress analysis method
Characteristic value, multiple characteristic of division values and multiple similarity characteristic values input the machine learning model and are trained.
Optionally, the score value computing module is in the following manner by the real-time subject information and the preset quantity
A history subject information inputs the machine learning model and each of described preset quantity history subject information is calculated
The score value of history subject information:
Obtain the label characteristics value, characteristic of division value and similarity characteristic value of the real-time subject information；
Label characteristics value, the classification for obtaining each history subject information in the preset quantity history subject information are special
Value indicative and similarity characteristic value；
By the label characteristics value of the real-time subject information, characteristic of division value and similarity characteristic value and the present count
Measure the label characteristics value, characteristic of division value and the input of similarity characteristic value of each history subject information in a history subject information
Each history subject information in the preset quantity history subject information is calculated in the machine learning model that training is completed
Score value.
The embodiment of the invention also provides a kind of server-side, including memory, processor and storage are on a memory simultaneously
The computer program that can be run on a processor, the processor realize above-mentioned intelligent answer when executing the computer program
Matching process.
The embodiment of the invention also provides a kind of computer readable storage medium, the readable storage medium storing program for executing includes computer
Program, the server-side computer program controls the readable storage medium storing program for executing when running where execute above-mentioned intelligent answer matching
Method.
Beneficial effect
Intelligent answer matching process and device provided in an embodiment of the present invention, be primarily based in presetting database all goes through
History subject information is trained the convolutional neural networks model of building, the convolutional neural networks model then finished by training
The second similarity value of each history subject information to match with real-time subject information is calculated, and is pressed based on the second similarity value
Preset quantity history theme letter is chosen from the multiple history subject informations to match with real-time subject information according to setting sequence
Breath calculates the score value to preset quantity history subject information using machine learning model, is finally carried out based on score value double
Major punishment is fixed to obtain the question and answer pair for meeting decision condition, in this way, can be real by convolutional neural networks model, machine learning model
Now to the multi-angular analysis of question information and processing, and then improve the quality of the question and answer pair acquired.
Further, it searches matched user according to real-time subject information to draw a portrait corresponding client, and by question information
It is sent to the client, improves the flexibility to question information processing, and the client can be guaranteed for the question information
The quality of given answer.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, below will be to needed in the embodiment attached
Figure is briefly described, it should be understood that the following drawings illustrates only certain embodiments of the present invention, therefore is not construed as pair
The restriction of range for those of ordinary skill in the art without creative efforts, can also be according to this
A little attached drawings obtain other relevant attached drawings.
Fig. 1 is a kind of block diagram of server-side 10 provided by the embodiment of the present invention.
Fig. 2 is a kind of flow chart of intelligent answer matching process provided by the embodiment of the present invention.
Fig. 3 is the schematic diagram for the sub-step that step S24 shown in Fig. 2 includes in an embodiment.
Fig. 4 is a kind of module frame chart of intelligent answer coalignment 20 provided by the embodiment of the present invention.
Icon:
10- server-side；11- memory；12- processor；13- network module；
20- intelligent answer coalignment；The real-time subject information extraction module of 21-；22- convolutional neural networks model construction mould
Block；23- subject information screening module；24- score value computing module；25- first judgment module；The second judgment module of 26-.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment only
It is a part of the embodiments of the present invention, instead of all the embodiments.The present invention being usually described and illustrated herein in the accompanying drawings
The component of embodiment can be arranged and be designed with a variety of different configurations.
Therefore, the detailed description of the embodiment of the present invention provided in the accompanying drawings is not intended to limit below claimed
The scope of the present invention, but be merely representative of selected embodiment of the invention.Based on the embodiments of the present invention, this field is common
Technical staff's every other embodiment obtained without creative efforts belongs to the model that the present invention protects
It encloses.
It should also be noted that similar label and letter indicate similar terms in following attached drawing, therefore, once a certain Xiang Yi
It is defined in a attached drawing, does not then need that it is further defined and explained in subsequent attached drawing.
Inventor further investigation reveals that, existing question answering system for user propose new problem processing mode it is single, flexibly
Property is not high, and then causes the answer matched of low quality.
Defect present in the above scheme in the prior art, is that inventor is obtaining after practicing and carefully studying
As a result, therefore, the solution that the discovery procedure of the above problem and the hereinafter embodiment of the present invention are proposed regarding to the issue above
Scheme all should be the contribution that inventor makes the present invention in process of the present invention.
Based on the studies above, the embodiment of the invention provides a kind of intelligent answer matching process and devices, pass through convolution mind
It can be realized multi-angular analysis and processing to question information through network model, machine learning model, and then guarantee to acquire
Question and answer pair quality.
Fig. 1 shows a kind of block diagram of server-side 10 provided by the embodiment of the present invention.In the embodiment of the present invention
Server-side 10 have data storage, transmission, processing function, as shown in Figure 1, server-side 10 includes: memory 11, processor
12, network module 13 and intelligent answer coalignment 20.
It is directly or indirectly electrically connected between memory 11, processor 12 and network module 13, to realize the biography of data
Defeated or interaction.It is electrically connected for example, these elements can be realized from each other by one or more communication bus or signal wire.
Intelligent answer coalignment 20 is stored in memory 11, the intelligent answer coalignment 20 includes at least one can be with software
Or the form of firmware (firmware) is stored in the software function module in the memory 11, the processor 12 passes through operation
The intelligent answer coalignment 20 being stored in the software program and module, such as the embodiment of the present invention in memory 11, from
And perform various functions application and data processing, that is, realize the intelligent answer matching process in the embodiment of the present invention.
Wherein, the memory 11 may be, but not limited to, random access memory (Random Access Memory,
RAM), read-only memory (Read Only Memory, ROM), programmable read only memory (Programmable Read-Only
Memory, PROM), erasable read-only memory (Erasable Programmable Read-Only Memory, EPROM),
Electricallyerasable ROM (EEROM) (Electric Erasable Programmable Read-Only Memory, EEPROM) etc..
Wherein, memory 11 is for storing program, and the processor 12 executes described program after receiving and executing instruction.
The processor 12 may be a kind of IC chip, the processing capacity with data.Above-mentioned processor 12
It can be general processor, including central processing unit (Central Processing Unit, CPU), network processing unit
(Network Processor, NP) etc..It may be implemented or execute each method, step disclosed in the embodiment of the present invention and patrol
Collect block diagram.General processor can be microprocessor or the processor is also possible to any conventional processor etc..
Network module 13 is used to establish the communication connection between server-side 10 and other communication terminal devices by network, real
The transmitting-receiving operation of existing network signal and data.Above-mentioned network signal may include wireless signal or wire signal.
It is appreciated that structure shown in FIG. 1 is only to illustrate, server-side 10 may also include than shown in Fig. 1 more or more
Few component, or with the configuration different from shown in Fig. 1.Each component shown in Fig. 1 can use hardware, software or its group
It closes and realizes.
The embodiment of the present invention also provides a kind of computer readable storage medium, and the readable storage medium storing program for executing includes computer journey
Sequence.Server-side 10 computer program controls the readable storage medium storing program for executing when running where executes following intelligent answer matching
Method.
Fig. 2 shows a kind of flow charts of intelligent answer matching process provided by the embodiment of the present invention.The method has
Method and step defined in the process of pass is applied to server-side 10, can be realized by the processor 12.It below will be to shown in Fig. 2
Detailed process be described in detail:
In the present embodiment, server-side 10 is connect with a client communication, and server-side 10 includes a presetting database, this is pre-
If being stored with multiple question and answer pair in database, these question and answer are to user's offer by 10 place platform of server-side, wherein Mei Gewen
It answers questions and is corresponding with a history subject information, for example, the history subject information in presetting database is M1~Mn。
Step S21 receives the question information that client is sent, extracts the real-time subject information of question information.
Server-side 10 carries out keyword extraction to the question information received, and then obtains the corresponding real-time master of question information
Inscribe information Mt, for question and answer matching later.
Step S22 constructs convolutional neural networks model, and all history subject informations in presetting database roll up input
Product neural network model is trained.
Convolutional neural networks (Convolutional Neural Networks, CNN) model is constructed, by M1~MnInput
CNN model is trained.
Step S23 finds out the multiple history subject informations to match with real-time subject information from presetting database, will
The convolutional neural networks that multiple history subject information input training are completed are gone through with calculating each of multiple history subject informations
The second similarity value between history subject information and real-time subject information is pressed based on multiple second similarity values being calculated
Preset quantity history subject information is chosen from multiple history subject informations according to setting sequence.
For convenient for subsequent explanation, it is assumed that with MtThe multiple history subject informations to match are M1~M20, wherein M1~M20In
Each history subject information and MtBetween the first similarity value be less than first threshold, first threshold can be according to the actual situation
Setting, the first similarity value are cosine similarity value.
Optionally, server-side 10 can use search engine and find out from presetting database and real-time subject information phase
The multiple history subject informations matched.
In the present embodiment, using presetting database as first database, the database that search engine is inquired as
Second database will complete the CNN model of training as third database.
By M1~M20Input training completes amount CNN model to calculate M1~M20In each history subject information and Mt
Between the second similarity value, multiple second similarity values being calculated are ranked up according to sequence from high to low, and
The corresponding history subject information of the second similarity value of forward setting quantity that sorts is obtained, for example, set quantity as five, acquisition
First five history subject information are as follows: M3、M1、M2、M8And M13。
Step S24, according to the building of multiple history subject informations and training machine learning model, by real-time subject information and in advance
If quantity history subject information input machine learning model is calculated each of preset quantity history subject information and goes through
The score value of history subject information.
Fig. 3 is please referred to, is enumerated in the present embodiment by step S241, step S242, step S243 and step S244
One of implementation of step S24.
Step S241 believes for each history theme in the multiple history subject informations to match with real-time subject information
Breath, obtains the label characteristics value, characteristic of division value and similarity characteristic value for including in the history subject information.
Extract M1~M20In each history subject information label characteristics value, characteristic of division value and similarity characteristic value.
Step S242 establishes building machine learning model based on gradient boosted tree and algorithm with regress analysis method, will acquire to obtain
Multiple label characteristics values, multiple characteristic of division values and multiple similarity characteristic values input machine learning model be trained.
Wherein, gradient boosted tree (Gradient Boosting Decision Tree, GBDT) has good non-linear
Capability of fitting and robustness, and algorithm with regress analysis method (Logistic Regression, LR) can be adapted for continuity and classification
Property independent variable, and be easy to use and explain.
It can be between history subject information and real-time subject information based on GBDT and the LR machine learning model built
Correlation is further analyzed and scores, and improves the accuracy and reliability of analysis.
Step S243 obtains the label characteristics value, characteristic of division value and similarity characteristic value of real-time subject information, obtains pre-
If label characteristics value, characteristic of division value and the similarity feature of each history subject information in quantity history subject information
Value.
M is obtained respectivelyt、M3、M1、M2、M8And M13Label characteristics value, characteristic of division value and similarity characteristic value.
Step S244 by the label characteristics value of real-time subject information, characteristic of division value and similarity characteristic value and is preset
The label characteristics value of each history subject information in quantity history subject information, characteristic of division value and similarity characteristic value are defeated
Enter the machine learning model that training is completed and each history subject information in preset quantity history subject information is calculated
Score value.
By Mt、M3、M1、M2、M8And M13Label characteristics value, characteristic of division value and similarity characteristic value input machine learning
Model calculates M3、M1、M2、M8And M13Respective score value.
Step S25, judges whether the maximum value in the multiple score values being calculated reaches setting value.
For example, M3、M1、M2、M8And M13Respective score value is respectively as follows: grade3、grade1、grade2、grade8With
grade13。
Assuming that maximum value is grade13If grade13Reach setting value (being assumed to be 80 points), turns to step S26.
If grade13Do not reach setting value, turns to step S29.
Step S26 calculates the editing distance value between the corresponding history subject information of maximum value and real-time subject information, sentences
Whether disconnected editing distance value is less than second threshold.
Calculate M13M betweentBetween editing distance value.
If editing distance value is less than second threshold (being set according to actual conditions), step S27 is turned to.
If editing distance value is not less than second threshold, step S28 is turned to.
Step S27 obtains the question and answer pair of the corresponding history subject information of maximum value, by question and answer to being sent to client.
Obtain M13Corresponding question and answer pair, by question and answer to being sent to client so that user checks.
Step S28 is modified acquisition correction result to maximum value, and correction result input machine learning model is carried out two
Secondary training.
It is appreciated that the machine learning model after can use second training is to M3、M1、M2、M8And M13It re-starts and comments
Score value calculates, then repeatedly step S25 and step S26.
Step S29 finds out client corresponding with user's portrait that real-time subject information matches, question information is sent out
It send to the client found out so that the client found out answers question information.
If grade13Do not reach setting value, shows M3、M1、M2、M8And M13The quality of mostly corresponding question and answer pair is difficult to and Mt
Matching, server-side 10 can be according to M at this timetLabel characteristics value, characteristic of division value etc. is found out and MtMatched user's portrait institute is right
The client answered is sent to the client found out by and by question information so that the client's single pair question information found out carries out
It answers.
In this way, being difficult to find out and M in server-side 10tThe high question and answer clock synchronization of matching precision, can be to the special of question information
The characteristics such as industry degree and region are analyzed, and then find out relatively suitable target user, determine question information by these mesh
It marks user to answer, to improve the professional and validity of question answering.
Optionally, the question information and answer form question and answer to being stored in first database, to realize to the first number
According to the update in library.
It is appreciated that first database, the second database and third database have the function of real-time update, so, it is possible
It improves to the subsequent accuracy for puing question to matching answer, and then guarantees the quality of the answer matched.
On the basis of the above, described as shown in figure 4, the embodiment of the invention provides a kind of intelligent answer coalignment 20
Intelligent answer coalignment 20 includes: real-time subject information extraction module 21, convolutional neural networks model construction module 22, theme
Information sifting module 23, score value computing module 24, first judgment module 25 and the second judgment module 26.
Real-time subject information extraction module 21, the question information sent for receiving the client extract the enquirement
The real-time subject information of information.
Since real-time subject information extraction module 21 is similar with the realization principle of step S21 in Fig. 2, do not make herein more
More explanations.
Convolutional neural networks model construction module 22 will be in presetting database for constructing convolutional neural networks model
All history subject informations are trained to the convolutional neural networks model is inputted.
Since convolutional neural networks model construction module 22 is similar with the realization principle of step S22 in Fig. 2, herein not
Make more explanations.
Subject information screening module 23, for being found out from the presetting database and the real-time subject information phase
The multiple history subject informations matched, wherein each history subject information in the multiple history subject information and it is described in real time
The first similarity value between subject information is less than first threshold；The multiple history subject information is inputted into the volume that training is completed
Product neural network model is to calculate each history subject information in the multiple history subject information and the real-time theme
The second similarity value between information；Based on multiple second similarity values being calculated, according to setting sequence from the multiple
Preset quantity history subject information is chosen in history subject information.
Since subject information screening module 23 is similar with the realization principle of step S23 in Fig. 2, do not say more herein
It is bright.
Score value computing module 24, for constructing simultaneously training machine learning model according to the multiple history subject information,
The real-time subject information and the preset quantity history subject information are inputted into the machine learning model, institute is calculated
State the score value of each history subject information in preset quantity history subject information.
Since score value computing module 24 is similar with the realization principle of step S24 in Fig. 2, do not say more herein
It is bright.
First judgment module 25, for judging whether the maximum value in the multiple score values being calculated reaches setting value,
If the maximum value reaches the setting value, the corresponding history subject information of the maximum value and the real-time subject information are calculated
Between editing distance value.
Since first judgment module 25 and step S25, step S26 in Fig. 2 are similar with the realization principle of step S29,
Do not illustrate more herein.
Second judgment module 26, for judging whether the editing distance value is less than second threshold, if the editing distance
Value is less than the second threshold, the question and answer pair of the corresponding history subject information of the maximum value is obtained, by the question and answer to transmission
To the client.
Since the second judgment module 26 and step S26, step S27 in Fig. 2 are similar with the realization principle of step S28,
Do not illustrate more herein.
To sum up, intelligent answer matching process and device provided by the embodiment of the present invention, by convolutional neural networks model,
Machine learning model can be realized multi-angular analysis and processing to question information, be matched by being searched according to real-time subject information
User draw a portrait corresponding client, and question information is sent to the client, improved to the flexible of question information processing
Property, it so, it is possible the quality for improving the question and answer pair acquired.
In several embodiments provided by the embodiment of the present invention, it should be understood that disclosed device and method, it can also
To realize by another way.Device and method embodiment described above is only schematical, for example, in attached drawing
Flow chart and block diagram show that the devices of multiple embodiments according to the present invention, method and computer program product are able to achieve
Architecture, function and operation.In this regard, each box in flowchart or block diagram can represent module, a program
A part of section or code, a part of the module, section or code include that one or more is patrolled for realizing defined
Collect the executable instruction of function.It should also be noted that in some implementations as replacement, function marked in the box
It can occur in a different order than that indicated in the drawings.For example, two continuous boxes can actually be held substantially in parallel
Row, they can also be executed in the opposite order sometimes, and this depends on the function involved.It is also noted that block diagram and/or
The combination of each box in flow chart and the box in block diagram and or flow chart, can the function as defined in executing or dynamic
The dedicated hardware based system made is realized, or can be realized using a combination of dedicated hardware and computer instructions.
In addition, each functional module in each embodiment of the present invention can integrate one independent portion of formation together
Point, it is also possible to modules individualism, an independent part can also be integrated to form with two or more modules.
It, can be with if the function is realized and when sold or used as an independent product in the form of software function module
It is stored in a computer readable storage medium.Based on this understanding, technical solution of the present invention is substantially in other words
The part of the part that contributes to existing technology or the technical solution can be embodied in the form of software products, the meter
Calculation machine software product is stored in a storage medium, including some instructions are used so that a computer equipment (can be a
People's computer, server-side 10 or the network equipment etc.) execute all or part of step of each embodiment the method for the present invention
Suddenly.And storage medium above-mentioned includes: USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), deposits at random
The various media that can store program code such as access to memory (RAM, Random Access Memory), magnetic or disk.
It should be noted that, in this document, the terms "include", "comprise" or its any other variant are intended to the packet of nonexcludability
Contain, so that the process, method, article or equipment for including a series of elements not only includes those elements, but also including
Other elements that are not explicitly listed, or further include for elements inherent to such a process, method, article, or device.
In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including the element
Process, method, article or equipment in there is also other identical elements.
The foregoing is only a preferred embodiment of the present invention, is not intended to restrict the invention, for the skill of this field
For art personnel, the invention may be variously modified and varied.All within the spirits and principles of the present invention, made any to repair
Change, equivalent replacement, improvement etc., should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of intelligent answer matching process, which is characterized in that applied to the server-side being connect with client communication, the method
Include:
The question information that the client is sent is received, the real-time subject information of the question information is extracted；
Convolutional neural networks model is constructed, by all history subject informations in presetting database to the input convolutional Neural net
Network model is trained；
The multiple history subject informations to match with the real-time subject information are found out from the presetting database, wherein
The first similarity value between each history subject information and the real-time subject information in the multiple history subject information
Less than first threshold；It is described to calculate that the multiple history subject information is inputted into the convolutional neural networks model that training is completed
The second similarity value between each history subject information and the real-time subject information in multiple history subject informations；It is based on
Multiple second similarity values being calculated choose preset quantity according to setting sequence from the multiple history subject information
History subject information；
According to the building of the multiple history subject information and training machine learning model, by the real-time subject information and described pre-
If quantity history subject information inputs the machine learning model and is calculated in the preset quantity history subject information
Each history subject information score value；
Judge whether the maximum value in the multiple score values being calculated reaches setting value, if the maximum value reaches the setting
Value, calculates the editing distance value between the corresponding history subject information of the maximum value and the real-time subject information；
Judge whether the editing distance value is less than second threshold, if the editing distance value is less than the second threshold, obtains
The question and answer pair of the corresponding history subject information of the maximum value, by the question and answer to being sent to the client.
2. intelligent answer matching process according to claim 1, which is characterized in that the method also includes:
If the maximum value does not reach the setting value, search whether there is the user to match with the real-time subject information
It draws a portrait corresponding client；If it exists, the question information is sent to the client found out so that the client found out
The question information is answered.
3. intelligent answer matching process according to claim 1, which is characterized in that based on multiple second phases being calculated
Like angle value, the step of choosing preset quantity history subject information from the multiple history subject information according to setting sequence,
Include:
Multiple second similarity values being calculated are ranked up according to sequence from high to low；
Obtain the corresponding history subject information of the second similarity value of the forward setting quantity that sorts.
4. intelligent answer matching process according to claim 1, which is characterized in that according to the multiple history subject information
The step of constructing simultaneously training machine learning model, comprising:
For each history subject information in the multiple history subject information, the mark for including in the history subject information is obtained
Sign characteristic value, characteristic of division value and similarity characteristic value；
Building machine learning model, the multiple label characteristics that will acquire are established based on gradient boosted tree and algorithm with regress analysis method
Value, multiple characteristic of division values and multiple similarity characteristic values input the machine learning model and are trained.
5. intelligent answer matching process according to claim 4, which is characterized in that by the real-time subject information and described
Preset quantity history subject information inputs the machine learning model and the preset quantity history subject information is calculated
In each history subject information score value the step of, comprising:
Obtain the label characteristics value, characteristic of division value and similarity characteristic value of the real-time subject information；
Obtain label characteristics value, the characteristic of division value of each history subject information in the preset quantity history subject information
With similarity characteristic value；
By the label characteristics value of the real-time subject information, characteristic of division value and similarity characteristic value and the preset quantity
Label characteristics value, characteristic of division value and the input training of similarity characteristic value of each history subject information in history subject information
Each history subject information that the machine learning model of completion is calculated in the preset quantity history subject information is commented
Score value.
6. a kind of intelligent answer coalignment, which is characterized in that applied to the server-side being connect with client communication, described device
Include:
Real-time subject information extraction module, the question information sent for receiving the client, extracts the question information
Real-time subject information；
Convolutional neural networks model construction module goes through all in presetting database for constructing convolutional neural networks model
History subject information is trained to the convolutional neural networks model is inputted；
Subject information screening module, for found out from the presetting database match with the real-time subject information it is more
A history subject information, wherein each history subject information and the real-time theme in the multiple history subject information are believed
The first similarity value between breath is less than first threshold；The multiple history subject information is inputted into the convolutional Neural that training is completed
Network model with calculate each history subject information in the multiple history subject information and the real-time subject information it
Between the second similarity value；Based on multiple second similarity values being calculated, according to setting sequence from the multiple history master
It inscribes and chooses preset quantity history subject information in information；
Score value computing module will be described for constructing simultaneously training machine learning model according to the multiple history subject information
Real-time subject information and the preset quantity history subject information input the machine learning model and described preset are calculated
The score value of each history subject information in quantity history subject information；
First judgment module, for judging whether the maximum value in the multiple score values being calculated reaches setting value, if described
Maximum value reaches the setting value, calculates between the corresponding history subject information of the maximum value and the real-time subject information
Editing distance value；
Second judgment module, for judging whether the editing distance value is less than second threshold, if the editing distance value is less than
The second threshold obtains the question and answer pair of the corresponding history subject information of the maximum value, and the question and answer are described to being sent to
Client.
7. intelligent answer coalignment according to claim 6, which is characterized in that the first judgment module is also used to:
If the maximum value does not reach the setting value, search whether there is the user to match with the real-time subject information
It draws a portrait corresponding client；If it exists, the question information is sent to the client found out so that the client found out
The question information is answered.
8. intelligent answer coalignment according to claim 6, which is characterized in that the subject information screening module passes through
Following manner is selected from the multiple history subject information based on multiple second similarity values being calculated, according to setting sequence
Take preset quantity history subject information:
Multiple second similarity values being calculated are ranked up according to sequence from high to low；
Obtain the corresponding history subject information of the second similarity value of the forward setting quantity that sorts.
9. intelligent answer coalignment according to claim 6, which is characterized in that the score value computing module by with
Under type is according to the building of the multiple history subject information and training machine learning model:
For each history subject information in the multiple history subject information, the mark for including in the history subject information is obtained
Sign characteristic value, characteristic of division value and similarity characteristic value；
Building machine learning model, the multiple label characteristics that will acquire are established based on gradient boosted tree and algorithm with regress analysis method
Value, multiple characteristic of division values and multiple similarity characteristic values input the machine learning model and are trained.
10. intelligent answer coalignment according to claim 9, which is characterized in that the score value computing module passes through
The real-time subject information and the preset quantity history subject information are inputted the machine learning model meter by following manner
It calculates and obtains the score value of each history subject information in the preset quantity history subject information:
Obtain the label characteristics value, characteristic of division value and similarity characteristic value of the real-time subject information；
Obtain label characteristics value, the characteristic of division value of each history subject information in the preset quantity history subject information
With similarity characteristic value；
By the label characteristics value of the real-time subject information, characteristic of division value and similarity characteristic value and the preset quantity
Label characteristics value, characteristic of division value and the input training of similarity characteristic value of each history subject information in history subject information
Each history subject information that the machine learning model of completion is calculated in the preset quantity history subject information is commented
Score value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811448785.3A CN109582773B (en) | 2018-11-29 | 2018-11-29 | Intelligent question-answer matching method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811448785.3A CN109582773B (en) | 2018-11-29 | 2018-11-29 | Intelligent question-answer matching method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109582773A true CN109582773A (en) | 2019-04-05 |
| CN109582773B CN109582773B (en) | 2020-11-27 |
Family
ID=65923775
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811448785.3A Active CN109582773B (en) | 2018-11-29 | 2018-11-29 | Intelligent question-answer matching method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109582773B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110096580A (en) * | 2019-04-24 | 2019-08-06 | 北京百度网讯科技有限公司 | A kind of FAQ dialogue method, device and electronic equipment |
| CN110347786A (en) * | 2019-06-11 | 2019-10-18 | 深圳追一科技有限公司 | A kind of tuning method and system of semantic model |
| CN110413961A (en) * | 2019-06-21 | 2019-11-05 | 平安国际智慧城市科技股份有限公司 | The method, apparatus and computer equipment of text scoring are carried out based on disaggregated model |
| CN111611363A (en) * | 2020-05-13 | 2020-09-01 | 湖南福米信息科技有限责任公司 | Intelligent response robot system based on user image and machine learning |
| CN112860962A (en) * | 2021-02-10 | 2021-05-28 | 北京百度网讯科技有限公司 | Method, device, electronic equipment and medium for processing question information |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103902733A (en) * | 2014-04-18 | 2014-07-02 | 北京大学 | Information retrieval method based on interrogative extension |
| US20160342895A1 (en) * | 2015-05-21 | 2016-11-24 | Baidu Usa Llc | Multilingual image question answering |
| CN106649760A (en) * | 2016-12-27 | 2017-05-10 | 北京百度网讯科技有限公司 | Question type search work searching method and question type search work searching device based on deep questions and answers |
| CN107145573A (en) * | 2017-05-05 | 2017-09-08 | 上海携程国际旅行社有限公司 | The problem of artificial intelligence customer service robot, answers method and system |
| CN107562836A (en) * | 2017-06-07 | 2018-01-09 | 北京航空航天大学 | Method is recommended based on the answerer of topic model and machine learning |
| CN108052588A (en) * | 2017-12-11 | 2018-05-18 | 浙江大学城市学院 | A kind of construction method of the document automatically request-answering system based on convolutional neural networks |
| CN108090174A (en) * | 2017-12-14 | 2018-05-29 | 北京邮电大学 | A kind of robot answer method and device based on system function syntax |
| CN108829719A (en) * | 2018-05-07 | 2018-11-16 | 中国科学院合肥物质科学研究院 | The non-true class quiz answers selection method of one kind and system |
-
2018
- 2018-11-29 CN CN201811448785.3A patent/CN109582773B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103902733A (en) * | 2014-04-18 | 2014-07-02 | 北京大学 | Information retrieval method based on interrogative extension |
| US20160342895A1 (en) * | 2015-05-21 | 2016-11-24 | Baidu Usa Llc | Multilingual image question answering |
| CN106649760A (en) * | 2016-12-27 | 2017-05-10 | 北京百度网讯科技有限公司 | Question type search work searching method and question type search work searching device based on deep questions and answers |
| CN107145573A (en) * | 2017-05-05 | 2017-09-08 | 上海携程国际旅行社有限公司 | The problem of artificial intelligence customer service robot, answers method and system |
| CN107562836A (en) * | 2017-06-07 | 2018-01-09 | 北京航空航天大学 | Method is recommended based on the answerer of topic model and machine learning |
| CN108052588A (en) * | 2017-12-11 | 2018-05-18 | 浙江大学城市学院 | A kind of construction method of the document automatically request-answering system based on convolutional neural networks |
| CN108090174A (en) * | 2017-12-14 | 2018-05-29 | 北京邮电大学 | A kind of robot answer method and device based on system function syntax |
| CN108829719A (en) * | 2018-05-07 | 2018-11-16 | 中国科学院合肥物质科学研究院 | The non-true class quiz answers selection method of one kind and system |
Non-Patent Citations (3)
| Title |
|---|
| RONGQIANG MA 等: "Hybrid answer selection model for non-factoid question answering", 《ASIAN LANGUAGE PROCESSING》 * |
| 张栋 等: "基于问题与答案联合表示学习的半监督问题分类方法", 《中文信息学报》 * |
| 荣光辉 等: "基于深度学习的问答匹配方法", 《计算机应用》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110096580A (en) * | 2019-04-24 | 2019-08-06 | 北京百度网讯科技有限公司 | A kind of FAQ dialogue method, device and electronic equipment |
| CN110347786A (en) * | 2019-06-11 | 2019-10-18 | 深圳追一科技有限公司 | A kind of tuning method and system of semantic model |
| CN110347786B (en) * | 2019-06-11 | 2021-01-05 | 深圳追一科技有限公司 | Semantic model tuning method and system |
| CN110413961A (en) * | 2019-06-21 | 2019-11-05 | 平安国际智慧城市科技股份有限公司 | The method, apparatus and computer equipment of text scoring are carried out based on disaggregated model |
| CN110413961B (en) * | 2019-06-21 | 2021-02-09 | 平安国际智慧城市科技股份有限公司 | Method and device for text scoring based on classification model and computer equipment |
| CN111611363A (en) * | 2020-05-13 | 2020-09-01 | 湖南福米信息科技有限责任公司 | Intelligent response robot system based on user image and machine learning |
| CN112860962A (en) * | 2021-02-10 | 2021-05-28 | 北京百度网讯科技有限公司 | Method, device, electronic equipment and medium for processing question information |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109582773B (en) | 2020-11-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109582773A (en) | Intelligent answer matching process and device | |
| CN109271401B (en) | Topic searching and correcting method and device, electronic equipment and storage medium | |
| Shovon et al. | An Approach of Improving Students Academic Performance by using k means clustering algorithm and Decision tree | |
| WO2017167069A1 (en) | Resume assessment method and apparatus | |
| CN106649760A (en) | Question type search work searching method and question type search work searching device based on deep questions and answers | |
| CN105117460A (en) | Learning resource recommendation method and system | |
| CN108121702B (en) | Method and system for evaluating and reading mathematical subjective questions | |
| CN105303179A (en) | Fingerprint identification method and fingerprint identification device | |
| CN109582969A (en) | Methodology for Entities Matching, device and electronic equipment | |
| CN108052504B (en) | Structure analysis method and system for mathematic subjective question answer result | |
| CN109273023B (en) | Data evaluation method, device and equipment and readable storage medium | |
| CN107301164B (en) | Semantic analysis method and device for mathematical formula | |
| CN103488790A (en) | Polychronic time sequence similarity analysis method based on weighting BORDA counting method | |
| CN112100464B (en) | Question-answering community expert recommendation method and system combining dynamic interest and professional knowledge | |
| CN115798022B (en) | Artificial intelligence recognition method based on feature extraction | |
| CN111611781B (en) | Data labeling method, question answering device and electronic equipment | |
| CN111597305B (en) | Entity marking method, entity marking device, computer equipment and storage medium | |
| CN112836994A (en) | Big data visualization method for online remote education and online education server | |
| CN110852064B (en) | Key sentence extraction method and device | |
| CN111541911A (en) | Video detection method and device, storage medium and electronic device | |
| CN109064578A (en) | A kind of attendance system and method based on cloud service | |
| CN114971425B (en) | Database information monitoring method, device, equipment and storage medium | |
| CN116228361A (en) | Course recommendation method, device, equipment and storage medium based on feature matching | |
| CN113313470B (en) | Employment type assessment method and system based on big data | |
| CN111708890B (en) | Search term determining method and related device |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |