CN107305578A - Human-machine intelligence's answering method and device - Google Patents
Human-machine intelligence's answering method and device Download PDFInfo
- Publication number
- CN107305578A CN107305578A CN201610262740.1A CN201610262740A CN107305578A CN 107305578 A CN107305578 A CN 107305578A CN 201610262740 A CN201610262740 A CN 201610262740A CN 107305578 A CN107305578 A CN 107305578A
- Authority
- CN
- China
- Prior art keywords
- answer
- language material
- question
- similarity
- corpus
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
This application discloses a kind of human-machine intelligence's answering method and device.One embodiment of methods described includes:The counseling problem of user's input is received, wherein, the counseling problem is made up of word;The Question-Answering Model that the counseling problem is inputted into training in advance obtains the degree of correlation of each answer language material in the counseling problem and the first question and answer corpus, wherein, the Question-Answering Model is used for the degree of correlation for determining the counseling problem and answer language material;Answer language material with maximum relation degree in the first question and answer corpus is sent to the user.The embodiment improves the accuracy rate of human-machine intelligence's question and answer by calculating the degree of correlation of counseling problem and answer language material.
Description
Technical field
The application is related to field of computer technology, and in particular to Internet technical field, especially relates to
And human-machine intelligence's answering method and device.
Background technology
With the development of human-machine intelligence's technology, increasing question answering system uses human-machine intelligence's skill
Art.The problem of existing human-machine intelligence's question answering system is typically by default question and answer corpus language
Material with user input counseling problem matched, with determine in question and answer corpus with user input
Counseling problem similarity highest the problem of, it is then that the corresponding answer language material of the problem language material is anti-
Feed user.
This method by calculating counseling problem and problem language material similarity can only be from literal upper general
The language material progress of the problem of in user's input consulting sentence and question and answer corpus similarity mode, and it is right
It is closer in semanteme, but the discrepant situation of word, can be because similarity is low can not be effective
Ground is matched.And prior art generally only considers that language material and user are defeated the problem of in question and answer corpus
The correlation of the counseling problem entered, the correlation without considering counseling problem and answer language material,
This can reduce the accuracy rate of human-machine intelligence's question and answer.
The content of the invention
The purpose of the application is to propose a kind of improved human-machine intelligence's answering method and device, come
Solve the technical problem that background section above is mentioned.
In a first aspect, this application provides a kind of human-machine intelligence's answering method, methods described includes:
The counseling problem of user's input is received, wherein, the counseling problem is made up of word;Will be described
The Question-Answering Model of counseling problem input training in advance obtains the counseling problem and the first question and answer language material
The degree of correlation of each answer language material in storehouse, wherein, the Question-Answering Model is used to determine that the consulting is asked
Topic and the degree of correlation of answer language material;To there is maximum relation degree in the first question and answer corpus
Answer language material is sent to the user.
In certain embodiments, in the question and answer mould that the counseling problem is inputted to training in advance
Type obtains the degree of correlation step of each answer language material in the counseling problem and the first question and answer corpus
Before rapid, methods described also includes:The second question and answer corpus is obtained, wherein, described second asks
Answering corpus includes multiple question and answer language materials pair, the problem of each question and answer language material is to by corresponding to language material
With answer language material composition;Based on each problem in the counseling problem and the second question and answer corpus
The degree of correlation of language material, to each question and answer language material in the second question and answer corpus to being ranked up;
According to the sequence, the problem of choosing predetermined number language material, and by described problem language material and therewith
Corresponding answer language material constitutes the first question and answer corpus.
In certain embodiments, methods described also includes:The step of setting up Question-Answering Model, including:
Obtain the second question and answer corpus;Extract each question and answer language material pair in the second question and answer corpus
Semantic feature and non-semantic feature, determine the question and answer language material centering problem language material and answer language
The similarity and described problem language material of the semantic feature of material and the non-semantic feature of answer language material
Similarity;The similarity of similarity and the non-semantic feature based on the semantic feature, instruction
Get the Question-Answering Model.
In certain embodiments, methods described also includes:Obtain each in second corpus
Problem language material and corresponding answer language material, generate new question and answer language material pair at random, and by institute
New question and answer language material is stated to the 3rd question and answer corpus of composition.
In certain embodiments, it is described to extract each question and answer language material pair in the second question and answer corpus
Semantic feature and non-semantic feature, determine the question and answer language material centering problem language material and answer language
The similarity and described problem language material of the semantic feature of material and the non-semantic feature of answer language material
Similarity, including:Determine the similarity of the semantic feature and the similarity of non-semantic feature
Step, including:The first problem language material and the first answer language material of the first question and answer language material pair are obtained,
Wherein, the first question and answer language material is to for the second question and answer corpus or the 3rd question and answer language
Expect any question and answer language material pair in storehouse;Determine the first problem language material and the first answer language
The length value of Longest Common Substring in material is used as the first similarity;Determine the first problem language
The literal similarity of material and the first answer language material is used as the second similarity;Determine described first
The N-gram of problem language material and the first answer language material matching degree is as third phase like degree;
Determine the angle of the term vector of the first problem language material and the term vector of the first answer language material
It is used as the 4th similarity;Determine first described in the theme feature vector sum of the first problem language material
The similarity of the theme feature vector of answer language material is used as the 5th similarity;Wherein, described first
Similarity, second similarity and the third phase are the first problem language material and institute like degree
State the similarity of the non-semantic feature of the first answer language material, the 4th similarity and the described 5th
Similarity is that the first problem language material is similar to the semantic feature of the first answer language material
Degree;The step of similarity based on the similarity for determining the semantic feature and non-semantic feature
Suddenly, each first question and answer language material in the second question and answer corpus and the 3rd question and answer corpus is determined
First similarity of centering first problem language material and the first answer language material, the second similarity, the 3rd
Similarity, the 4th similarity and the 5th similarity.
In certain embodiments, the theme feature vector sum for determining the first problem language material
The first answer language material theme feature vector similarity as the 5th similarity, including:
According to each question and answer language material pair in the second question and answer corpus and/or the 3rd question and answer corpus
The problem of the corresponding theme of the language material and corresponding theme of answer language material, generate subject analysis model,
Wherein, the subject analysis model is used for any one problem language material or any one answer language
Material is converted into the theme feature vector of described problem language material or the answer language material;By described first
Problem and first answer input subject analysis model obtain the corresponding first problem language material
Theme feature vector sum described in the first answer language material theme feature vector;Described first is asked
Inscribe the similarity of the theme feature vector of the first answer language material described in the theme feature vector sum of language material
It is used as the 5th similarity.
Second aspect, this application provides a kind of human-machine intelligence's question and answer system, described device includes:
Counseling problem receiving module, is configured to receive the counseling problem of user's input, wherein, it is described
Counseling problem is made up of word;Degree of correlation acquisition module, is configured to the counseling problem is defeated
The Question-Answering Model for entering training in advance obtains each answer in the counseling problem and the first question and answer corpus
The degree of correlation of language material, wherein, the Question-Answering Model is used to determine the counseling problem and answer language
The degree of correlation of material;Answer language material sending module, is configured in the first question and answer corpus
Answer language material with maximum relation degree is sent to the user.
In certain embodiments, described device also includes:First question and answer corpus comprising modules,
It is configured to obtain the second question and answer corpus, wherein, the second question and answer corpus includes multiple
Question and answer language material pair, the problem of each question and answer language material is to by corresponding to language material and answer language material are constituted;
It is right based on the degree of correlation of each problem language material in the counseling problem and the second question and answer corpus
Each question and answer language material in the second question and answer corpus is to being ranked up;According to the sequence, choosing
The problem of taking predetermined number language material, and by described problem language material and corresponding answer language material group
Into the first question and answer corpus.
In certain embodiments, described device also includes:Question-Answering Model sets up module, and configuration is used
In setting up Question-Answering Model, including:Second question and answer corpus acquiring unit, is configured to obtain institute
State the second question and answer corpus;Similarity determining unit, is configured to extract the second question and answer language
Expect the semantic feature and non-semantic feature of each question and answer language material pair in storehouse, determine the question and answer language material pair
The similarity of the semantic feature of middle problem language material and answer language material and described problem language material and answer
The similarity of the non-semantic feature of language material;Question-Answering Model training unit, is configured to based on described
The similarity of the similarity of semantic feature and the non-semantic feature, training obtains the question and answer mould
Type.
In certain embodiments, described device also includes:3rd question and answer corpus comprising modules,
It is configured to obtain each problem language material and corresponding answer language in second corpus
Material, generates new question and answer language material pair, and the new question and answer language material is asked composition the 3rd at random
Answer corpus.
In certain embodiments, similarity determining unit configuration specifically for:It is determined that described
The step of similarity of the similarity of semantic feature and non-semantic feature, including:First is obtained to ask
The first problem language material and the first answer language material of language material pair are answered, wherein, the first question and answer language material
To for any question and answer language material in the second question and answer corpus or the 3rd question and answer corpus
It is right;Determine the first problem language material and the Longest Common Substring in the first answer language material
Length value is used as the first similarity;Determine the first problem language material and the first answer language material
Literal similarity be used as the second similarity;Determine that the first problem language material and described first is answered
The N-gram of case language material matching degree is as third phase like degree;Determine the first problem language material
Term vector and the angle of term vector of the first answer language material be used as the 4th similarity;It is determined that
The theme feature of first answer language material described in the theme feature vector sum of the first problem language material to
The similarity of amount is used as the 5th similarity;Wherein, first similarity, described second similar
Degree and the third phase are like the non-language spent for the first problem language material with the first answer language material
The similarity of adopted feature, the 4th similarity and the 5th similarity are the first problem
Language material and the similarity of the semantic feature of the first answer language material;Based on determination institute predicate
The step of similarity of the similarity of adopted feature and non-semantic feature, determine the second question and answer language
Expect each first question and answer language material centering first problem language material and first in storehouse and the 3rd question and answer corpus
First similarity of answer language material, the second similarity, third phase are like spending, the 4th similarity and the
Five similarities.
In certain embodiments, the similarity determining unit configuration is further used for:According to institute
State in the second question and answer corpus and/or the 3rd question and answer corpus each question and answer language material to the problem of
The corresponding theme of language material and the corresponding theme of answer language material, generate subject analysis model, wherein,
The subject analysis model is used to change any one problem language material or any one answer language material
Into the theme feature vector of described problem language material or the answer language material;By the first problem and
The first answer input subject analysis model obtains the theme of the corresponding first problem language material
The theme feature vector of characteristic vector and the first answer language material;By the first problem language material
Theme feature vector sum described in the similarity of theme feature vector of the first answer language material be used as the
Five similarities.
Human-machine intelligence's answering method and device that the application is provided, receive the official communication of user's input first
Inquiry is inscribed, and the counseling problem is inputted to the Question-Answering Model of training in advance afterwards, to obtain the official communication
The degree of correlation with each answer language material in the first question and answer corpus is inscribed in inquiry, will finally have maximum
The answer language material of the degree of correlation is sent to user, and methods described is by calculating counseling problem and answer language
The degree of correlation of material improves the accuracy rate of human-machine intelligence's question and answer.
Brief description of the drawings
Retouched by reading with reference to the detailed of being made to non-limiting example of being made of the following drawings
State, other features, objects and advantages will become more apparent upon:
Fig. 1 is that the application can apply to exemplary system architecture figure therein;
Fig. 2 is the flow chart of one embodiment of human-machine intelligence's answering method according to the application;
During Fig. 3 is human-machine intelligence's answering method according to the application, the question and answer of training in advance are obtained
A kind of indicative flowchart of implementation of model;
Fig. 4 is the flow of another embodiment of human-machine intelligence's answering method according to the application
Figure;
Fig. 5 is the structural representation of one embodiment of human-machine intelligence's question and answer system according to the application
Figure;
Fig. 6 is adapted for for realizing the terminal device of the embodiment of the present application or the computer of server
The structural representation of system.
Embodiment
The application is described in further detail with reference to the accompanying drawings and examples.It is appreciated that
, specific embodiment described herein is used only for explaining related invention, rather than to the hair
Bright restriction.It also should be noted that, illustrate only for the ease of description, in accompanying drawing with
About the related part of invention.
It should be noted that in the case where not conflicting, embodiment and embodiment in the application
In feature can be mutually combined.Describe this in detail below with reference to the accompanying drawings and in conjunction with the embodiments
Application.
Fig. 1, which is shown, can apply the human-machine intelligence's answering method or human-machine intelligence's question and answer of the application
The exemplary system architecture 100 of the embodiment of device.
As shown in figure 1, system architecture 100 can include terminal device 101,102,103,
Network 104 and server 105.Network 104 is used in the and of terminal device 101,102,103
The medium of communication link is provided between server 105.Network 104 can include various connection classes
Type, such as wired, wireless communication link or fiber optic cables etc..
User can pass through network 104 and server 105 with using terminal equipment 101,102,103
Interaction, to receive or send message etc..It can be provided with terminal device 101,102,103
Various telecommunication customer end applications, such as instant communication software, shopping class are applied, searching class is applied,
Web browser applications, social platform software etc..
Terminal device 101,102,103 can be with display screen and support human-machine intelligence to ask
The various electronic equipments answered, including but not limited to smart mobile phone, tablet personal computer, e-book reading
Device, MP3 player (Moving Picture Experts Group Audio Layer III, dynamic shadow
As expert's compression standard audio aspect 3), MP4 (Moving Picture Experts Group Audio
Layer IV, dynamic image expert's compression standard audio aspect 4) player, portable meter on knee
Calculation machine and desktop computer etc..
Server 105 can be to provide the server of various services, for example to terminal device 101,
102nd, 103 counseling problems sent provide the background server supported.Background server can be right
The data such as the counseling problem received such as are counted, analyzed at the processing, and result is fed back
To terminal device.
It should be noted that human-machine intelligence's answering method for being provided of the embodiment of the present application it is general by
Server 105 is performed, correspondingly, and human-machine intelligence's question and answer system is generally positioned at server 105
In.
It should be understood that the number of the terminal device, network and server in Fig. 1 is only signal
Property.According to needs are realized, can have any number of terminal device, network and server.
With continued reference to Fig. 2, it illustrates one of human-machine intelligence's answering method according to the application
The flow 200 of embodiment.Described human-machine intelligence's answering method, comprises the following steps:
Step 201, the counseling problem of user's input is received.
In the present embodiment, human-machine intelligence's answering method operation electronic equipment thereon is (for example
Server shown in Fig. 1) can be by wired connection mode or radio connection from user
The counseling problem of user's input is received using its terminal for carrying out human-machine intelligence's question and answer.Here, use
The counseling problem of family input is typically what is be made up of word, such as Chinese character, punctuate, numeral.
It is pointed out that above-mentioned radio connection can include but is not limited to 3G/4G connections,
WiFi connections, bluetooth connection, WiMAX connections, Zigbee connections, UWB (ultra wideband)
Connection and other currently known or exploitation in the future radio connections.
Step 202, the Question-Answering Model for counseling problem being inputted into training in advance obtains the counseling problem
With the degree of correlation of each answer language material in the first question and answer corpus.
In the present embodiment, the electronic equipment of above-mentioned human-machine intelligence's answering method operation thereon can
To obtain the first question and answer corpus in advance, the first question and answer corpus at least includes multiple answer languages
Material.And above-mentioned electronic equipment can be used with one Question-Answering Model of training in advance, the Question-Answering Model
In it is determined that the counseling problem and each answer language material in above-mentioned first question and answer corpus of user's input
The degree of correlation.The counseling problem of the user's input received based on step 201, above-mentioned electronic equipment
The counseling problem is inputted into above-mentioned Question-Answering Model, the counseling problem and above-mentioned first question and answer language is obtained
Expect the degree of correlation of each answer language material in storehouse.
In the present embodiment, the degree of correlation of above-mentioned counseling problem and answer language material can use a variety of shapes
Formula represents, including but not limited to form of percents or numerical values recited etc., then, by phase
Order descending Guan Du is ranked up to the above-mentioned degree of correlation, is pre-set in order to basis
Rule chooses the answer language material finally needed.It should be noted that above-mentioned consulting can be used here
The semantic similarity of question and answer language material characterizes the degree of correlation between the two, can also be with above-mentioned
The degree of correlation of the non-semantic similarity characterization of counseling problem and answer language material between the two, or will
The semantic similarity and non-semantic similarity of above-mentioned counseling problem and answer language material are characterized after blending
The degree of correlation between the two.
Step 203, the answer language material with maximum relation degree in the first question and answer corpus is sent
To user.
In the present embodiment, the above-mentioned counseling problem and the first corpus obtained based on step 202
In each answer language material the degree of correlation, the operation of human-machine intelligence's answering method can with electronic equipment thereon
To determine maximum relation degree in the above-mentioned degree of correlation, then obtain that the maximum relation degree is corresponding to be answered
Case language material, now it is considered that the answer language material is most can be accurate in above-mentioned first question and answer corpus
Answer the answer language material of the counseling problem of user's input.
Human-machine intelligence's answering method that above-described embodiment of the application is provided, receives user defeated first
The counseling problem entered, the counseling problem afterwards inputted the Question-Answering Model of training in advance to obtain this
The degree of correlation of each answer language material in counseling problem and the first question and answer corpus, will finally have most
The answer language material of the big degree of correlation is sent to user, and human-machine intelligence's answering method is seeked advice from by calculating
The degree of correlation of question and answer language material improves the accuracy rate of human-machine intelligence's question and answer.
In some optional schemes, the Question-Answering Model for the training in advance used in step 202 can
To be set up by flow 300 as shown in Figure 3.
Step 301, the second question and answer corpus is obtained.
Generally, in human-machine intelligence's question answering system, after user inputs a counseling problem, people
Machine intelligent Answer System can feed back an answer, form the question and answer language material pair of question-response.Therefore,
In this implementation, human-machine intelligence's answering method operation electronic equipment thereon can be from going through
The question and answer language material of above-mentioned question-response is obtained in nan-machine interrogation's data of history to the second question and answer language of composition
Expect storehouse.Here multiple question and answer language materials pair, each question and answer language can be included in the second question and answer corpus
The problem of material is to by corresponding to language material and answer language material are constituted.With asking for the question answering system in electric business field
Data instance is answered, its off-line data that can choose question-response constitutes above-mentioned second question and answer language material
Storehouse, and one can not be chosen ask and more answer or ask that one answers line advisory data more.Chosen by this mode
Question and answer data need not do the processing of question and answer pair again, can directly use.
It should be noted that above-mentioned the second question and answer language material directly obtained may include many classes
Like the garbage of " support for thanking you ", " wish you do shopping happiness " etc., this can influence training
Question-Answering Model effect.Therefore, above-mentioned electronic equipment can also be to above-mentioned second question and answer language material
Question and answer language material in storehouse carries out data cleansing, and above-mentioned garbage is deleted.
Step 302, the semantic feature of each question and answer language material pair is extracted in the second question and answer corpus and non-
Semantic feature, determine question and answer language material centering problem language material and answer language material semantic feature it is similar
The similarity of the non-semantic feature of degree and problem language material and answer language material.
In this implementation, the second question and answer corpus obtained based on step 301, above-mentioned electricity
Sub- equipment can obtain each question and answer language material pair in above-mentioned second question and answer corpus, can enter afterwards
One step obtains the semantic feature and non-semantic feature of above-mentioned each question and answer language material pair, to pass through various hands
Section obtain the semantic feature of each question and answer language material centering problem language material and answer language material similarity and
The similarity of the non-semantic feature of problem language material and answer language material.
In some optional implementations of the present embodiment, above-mentioned electronic equipment can be obtained
Each problem language material and corresponding answer language material in the second corpus are stated, for generating at random
New question and answer language material pair, and by the new question and answer language material of generation to the 3rd question and answer corpus of composition.
It is understood that the answer language material of each question and answer language material centering in above-mentioned second question and answer corpus
It is the related answers of correspondence problem language material, you can be considered positive example, and the 3rd of above-mentioned generation
The answer language material of the question and answer language material pair constituted at random in question and answer corpus and corresponding problem language
Material is typically unrelated answers, you can be considered negative example.Above-mentioned second question and answer corpus and the 3rd
Question and answer corpus may be used to train above-mentioned Question-Answering Model.
Generally, above-mentioned electronic equipment can be from many aspects such as literal similarity, Topic Similarities
Determine semantic feature similarity and the non-semantic spy of question and answer language material centering problem language material and answer language material
Levy similarity.Therefore, above-mentioned question and answer language can be determined as follows in this implementation
The problem of expecting centering language material is similar with non-semantic feature with the semantic feature similarity of answer language material
Degree:First, above-mentioned electronic equipment can obtain the first question and answer language material pair first problem language material and
First answer language material, the first question and answer language material here is to for above-mentioned second question and answer corpus or the 3rd
Any question and answer language material pair in question and answer corpus;Secondly, above-mentioned electronic equipment can determine above-mentioned
The length value of first problem language material and the Longest Common Substring in the first answer language material, by the length
It is worth the first similarity as above-mentioned first problem language material and the first answer language material, this is first similar
Spend the similarity of the non-semantic feature for first problem language material and the first answer language material;Afterwards, on
Stating electronic equipment can continue to determine the literal phase of above-mentioned first problem language material and the first answer language material
Like second similarity of the degree as above-mentioned first problem language material and the first answer language material, second phase
It is first problem language material and the similarity of the non-semantic feature of the first answer language material like degree;Then,
Above-mentioned electronic equipment can determine the N-gram of above-mentioned first problem language material and the first answer language material
Matching degree as the third phase of above-mentioned first problem language material and the first answer language material like spending, this
Three similarities are the similarity of the non-semantic feature of first problem language material and the first answer language material;Again
Person, above-mentioned electronic equipment can also determine above-mentioned first problem language material by methods such as neutral nets
Term vector and the first answer language material term vector, then calculate first problem language material term vector and
Vector angle between the term vector of first answer language material is used as above-mentioned first problem language material and first
4th similarity of answer language material, the 4th similarity is first problem language material and the first answer language
The similarity of the semantic feature of material;Finally, above-mentioned electronic equipment can also determine that above-mentioned first asks
The theme feature vector of theme feature vector sum the first answer language material of language material is inscribed, calculates both it
Between vector angle as above-mentioned first problem language material and the first answer language material the 5th similarity,
5th similarity is the similarity of the semantic feature of first problem language material and the first answer language material.
In this implementation, above-mentioned electronic equipment can determine above-mentioned second using above-mentioned steps
Each question and answer language material centering problem language material and answer language in question and answer corpus and the 3rd question and answer corpus
First similarity of material, the second similarity, third phase are similar like degree, the 4th similarity and the 5th
Degree, is used as the semantic feature similarity and non-semantic characteristic similarity of each question and answer language material pair.
In this implementation, above-mentioned electronic equipment can according to above-mentioned second question and answer corpus and/
Or the 3rd each question and answer language material in question and answer corpus to the problem of the corresponding theme of language material and answer language material
Corresponding theme, generates subject analysis model.Here subject analysis model can be used for appointing
One problem language material of meaning or any one answer language material are converted into above mentioned problem language material or answer language material
Theme feature vector.Afterwards, above-mentioned electronic equipment can answer above-mentioned first problem and first
Case inputs the theme feature vector sum that above-mentioned subject analysis model obtains corresponding first problem language material
The theme feature vector of first answer language material.Finally, above-mentioned electronic equipment can determine above-mentioned
Theme feature vector sum the first answer language material of one problem language material theme feature vector it is similar
Degree, the similarity is above-mentioned 5th similarity.
It should be noted that determining the semantic feature similarity of problem language material and answer language material and non-
The method not limited to this of semantic feature similarity, those skilled in the art can use others side
Formula determines above-mentioned semantic feature similarity and non-semantic characteristic similarity, and this is widely studied at present
With the known technology of application, it will not be repeated here.
Step 303, similarity and the similarity of non-semantic feature based on semantic feature, training
Obtain Question-Answering Model.
In this implementation, the second question and answer corpus and the 3rd obtained based on step 302 is asked
Answer each question and answer language material in corpus to the problem of language material and answer language material semantic feature it is similar
The similarity of degree and non-semantic feature, above-mentioned electronic equipment can be using the instruction such as linear regression algorithm
The similarity of the similarity and non-semantic feature of practicing above-mentioned each semantic feature generates above-mentioned question and answer mould
Type.Here Question-Answering Model can be shape such as y=f (x1,x2,x3...xn) formula, wherein, x1, x2,
x3…xnRefer to the similarity of above-mentioned semantic feature and/or the similarity of non-semantic feature respectively.
The training step for the Question-Answering Model that the implementation of above-described embodiment of the application is provided, leads to
Cross acquisition the second question and answer corpus and the 3rd question and answer corpus in each question and answer language material to the problem of language material
With the similarity and the similarity of non-semantic feature of multiple semantic features of answer language material, using line
Property the above-mentioned semantic feature of the training such as regression algorithm similarity and non-semantic feature similarity generation
Question-Answering Model, the model can calculate user's input counseling problem and each answer language material exactly
The degree of correlation, improve the accuracy rate of human-machine intelligence's question and answer.
With further reference to Fig. 4, it illustrates another embodiment of human-machine intelligence's answering method
Flow 400.The flow 400 of human-machine intelligence's answering method, comprises the following steps:
Step 401, the counseling problem of user's input is received.
In the present embodiment, human-machine intelligence's answering method operation electronic equipment thereon is (for example
Server shown in Fig. 1) can be by wired connection mode or radio connection from user
The counseling problem of user's input is received using its terminal for carrying out human-machine intelligence's question and answer.Here, use
The counseling problem of family input is typically what is be made up of word, such as Chinese character, punctuate, numeral.
Step 402, the second question and answer corpus is obtained.
In the present embodiment, above-mentioned electronic equipment can be obtained from nan-machine interrogation's data of history
The question and answer language material of question-response is to the second question and answer corpus of composition.Here the second question and answer corpus
In can include multiple question and answer language materials pair, each question and answer language material to by correspondence the problem of language material and answer
Language material is constituted.With the question and answer data instance of the question answering system in electric business field, it can choose one and ask
One off-line data answered constitutes above-mentioned second question and answer corpus, and can not choose one and ask answer or many more
Ask that one answers line advisory data.The question and answer data chosen by this mode need not do question and answer pair again
Processing, can directly be used.Step 403, based in counseling problem and the second question and answer corpus
The degree of correlation of each problem language material, to each question and answer language material in the second question and answer corpus to being ranked up.
In the present embodiment, above-mentioned electronic equipment can obtain above-mentioned consulting using lightweight algorithm
Problem and the degree of correlation of each problem language material in the second question and answer corpus, afterwards can be according to the degree of correlation
Order from big to small is by each question and answer language material in the second question and answer corpus to tentatively being sorted.
In some optional implementations of the present embodiment, above-mentioned lightweight algorithm can be example
Such as Lucene sort algorithm, such a algorithm comparison is simple, accuracy rate is relatively low, but calculates
Speed is fast.Therefore, here can be using above-mentioned algorithm to each question and answer in the second question and answer corpus
Language material according to order from big to small to tentatively being sorted, afterwards, then selects sequence
Forward multiple question and answer language materials are to carrying out further relatedness computation.
Step 404, according to sequence, the problem of choosing predetermined number language material, and by the problem language
Material and corresponding answer language material constitute the first question and answer corpus.
In the present embodiment, based in step 404 to each question and answer in above-mentioned second question and answer corpus
The sequence of language material, above-mentioned electronic equipment can choose the question and answer language material for the forward predetermined number that sorts
To constituting above-mentioned first question and answer corpus.The first question and answer corpus is accurately calculated for next step
The degree of correlation of above-mentioned counseling problem and answer language material.
Step 405, the Question-Answering Model for counseling problem being inputted into training in advance obtains the counseling problem
With the degree of correlation of each answer language material in the first question and answer corpus.
In the present embodiment, the first question and answer corpus obtained based on step 404, above-mentioned electronics
The counseling problem that equipment inputs the user received inputs above-mentioned Question-Answering Model, obtains the consulting
Problem and the degree of correlation of each answer language material in the first question and answer corpus.
Step 406, the answer language material with maximum relation degree in the first question and answer corpus is sent
To user.
In the present embodiment, the above-mentioned counseling problem and the first corpus obtained based on step 405
In each answer language material the degree of correlation, the operation of human-machine intelligence's answering method can with electronic equipment thereon
To determine maximum relation degree in the above-mentioned degree of correlation, then obtain that the maximum relation degree is corresponding to be answered
Case language material, now it is considered that the answer language material is most can be accurate in above-mentioned first question and answer corpus
Answer the answer language material of the counseling problem of user's input.
Figure 4, it is seen that compared with the corresponding embodiments of Fig. 2, the people in the present embodiment
The step of flow 400 of quick-witted energy answering method highlights the first question and answer corpus of acquisition.This reality
The predetermined number for having larger correlation with counseling problem can be obtained and ask by applying the scheme of example description
First question and answer corpus of the language material to composition is answered, so as to ensure that human-machine intelligence's question and answer are accurate
The arithmetic speed of Question-Answering Model is further improved while rate.
With further reference to Fig. 5, as the realization to method shown in above-mentioned each figure, the application is provided
A kind of one embodiment of human-machine intelligence's question and answer system, the device embodiment with shown in Fig. 2
Embodiment of the method is corresponding, and the device specifically can apply in various electronic equipments.
As shown in figure 5, human-machine intelligence's question and answer system 500 described in the present embodiment includes:Consulting
Problem receiving module 501, degree of correlation acquisition module 502 and answer language material sending module 503.Its
In, counseling problem receiving module 501 is configured to receive the counseling problem of user's input, wherein,
The counseling problem is made up of word;Degree of correlation acquisition module 502 is configured to ask above-mentioned consulting
The Question-Answering Model of topic input training in advance obtains each answer in counseling problem and the first question and answer corpus
The degree of correlation of language material, wherein, above-mentioned Question-Answering Model is used to determine above-mentioned counseling problem and answer language
The degree of correlation of material;Answer language material sending module 503 is configured to above-mentioned first question and answer corpus
In have maximum relation degree answer language material be sent to above-mentioned user.
In some optional implementations of the present embodiment, said apparatus 500 also includes first
Question and answer corpus comprising modules (not shown), is configured to obtain the second question and answer corpus, wherein,
Above-mentioned second question and answer corpus includes multiple question and answer language materials pair, and each question and answer language material by corresponding to being asked
Inscribe language material and answer language material composition;Based in above-mentioned counseling problem and above-mentioned second question and answer corpus
The degree of correlation of each problem language material, to each question and answer language material in the second question and answer corpus to arranging
Sequence;According to above-mentioned sequence, the problem of choosing predetermined number language material, and by the problem language material and with
Corresponding answer language material constitute above-mentioned first question and answer corpus.
In some optional implementations of the present embodiment, said apparatus 500 also includes question and answer
Model building module (not shown), is configured to set up Question-Answering Model, including:Second question and answer language
Expect storehouse acquiring unit (not shown), be configured to obtain above-mentioned second question and answer corpus;Similarity
Determining unit (not shown), is configured to extract each question and answer language material in above-mentioned second question and answer corpus
To semantic feature and non-semantic feature, determine question and answer language material centering problem language material and answer language material
Semantic feature similarity and problem language material and answer language material non-semantic feature it is similar
Degree;Question-Answering Model training unit (not shown), is configured to based on the similar of above-mentioned semantic feature
The similarity of degree and non-semantic feature, training obtains above-mentioned Question-Answering Model.
In some optional implementations of the present embodiment, said apparatus 500 also includes:The
Three question and answer corpus comprising modules (not shown), are configured to obtain in above-mentioned second corpus
Each problem language material and corresponding answer language material, generate new question and answer language material pair at random, and will
New question and answer language material is to the 3rd question and answer corpus of composition.
In some optional implementations of the present embodiment, above-mentioned similarity determining unit is (not
Show) configuration specifically for:Determine the similarity of above-mentioned semantic feature and the phase of non-semantic feature
The step of seemingly spending, including:Obtain first problem language material and the first answer of the first question and answer language material pair
Language material, wherein, the first question and answer language material is to for above-mentioned second question and answer corpus or the 3rd question and answer language material
Any question and answer language material pair in storehouse;Determine in above-mentioned first problem language material and the first answer language material
The length value of Longest Common Substring is used as the first similarity;Determine above-mentioned first problem language material and
The literal similarity of one answer language material is used as the second similarity;Determine above-mentioned first problem language material and
The N-gram of first answer language material matching degree is as third phase like degree;Determine that above-mentioned first asks
The angle for inscribing the term vector of language material and the term vector of the first answer language material is used as the 4th similarity;Really
The theme feature vector of theme feature vector sum the first answer language material of fixed above-mentioned first problem language material
Similarity be used as the 5th similarity;Wherein, above-mentioned first similarity, the second similarity and
Three similarities are the similarity of the non-semantic feature of first problem language material and the first answer language material, on
The semanteme that the 4th similarity and the 5th similarity are stated for first problem language material and the first answer language material is special
The similarity levied;The similarity of similarity and non-semantic feature based on above-mentioned determination semantic feature
The step of, determine each first question and answer in above-mentioned second question and answer corpus and the 3rd question and answer corpus
First similarity of language material centering first problem language material and the first answer language material, the second similarity,
Third phase is like degree, the 4th similarity and the 5th similarity.
In some optional implementations of the present embodiment, above-mentioned similarity determining unit is (not
Show) configuration be further used for:According to above-mentioned second question and answer corpus and/or the 3rd question and answer language material
In storehouse each question and answer language material to the problem of the corresponding theme of the language material and corresponding theme of answer language material, it is raw
Into subject analysis model, wherein, subject analysis model is used for any one problem language material or appointed
One answer language material of meaning is converted into the theme feature vector of the problem language material or answer language material;Will be upper
State first problem and the first answer input subject analysis model obtains corresponding first problem language material
The theme feature vector of theme feature vector sum the first answer language material;By the first problem language material
The similarity of the theme feature vector of theme feature vector sum the first answer language material is similar as the 5th
Degree.
It will be understood by those skilled in the art that above-mentioned human-machine intelligence's question and answer system 500 also includes one
Other a little known features, such as processor, memory, in order to unnecessarily obscure the disclosure
Embodiment, these known structures are not shown in Figure 5.
Below with reference to Fig. 6, it illustrates suitable for for realizing the terminal device of the embodiment of the present application
Or the structural representation of the computer system 600 of server.
As shown in fig. 6, computer system 600 includes CPU (CPU) 601, its
Can according to the program being stored in read-only storage (ROM) 602 or from storage part 608
The program that is loaded into random access storage device (RAM) 603 and perform various appropriate actions
And processing.In RAM 603, the system that is also stored with 600 operates required various program sums
According to.CPU 601, ROM 602 and RAM 603 are connected with each other by bus 604.Input
/ output (I/O) interface 605 is also connected to bus 604.
I/O interfaces 605 are connected to lower component:Importation 606 including keyboard, mouse etc.;
Including cathode-ray tube (CRT), liquid crystal display (LCD) etc. and loudspeaker etc.
Output par, c 607;Storage part 608 including hard disk etc.;And including such as LAN card,
The communications portion 609 of the NIC of modem etc..Communications portion 609 is via such as
The network of internet performs communication process.Driver 610 is also according to needing to be connected to I/O interfaces
605.Detachable media 611, such as disk, CD, magneto-optic disk, semiconductor memory etc.,
Be arranged on as needed on driver 610, in order to the computer program that reads from it according to
Need to be mounted into storage part 608.
Especially, in accordance with an embodiment of the present disclosure, the process described above with reference to flow chart can be with
It is implemented as computer software programs.For example, embodiment of the disclosure includes a kind of computer journey
Sequence product, it includes being tangibly embodied in the computer program on machine readable media, the meter
Calculation machine program bag, which contains, is used for the program code of the method shown in execution flow chart.Implement such
In example, the computer program can be downloaded and installed by communications portion 609 from network,
And/or be mounted from detachable media 611.
Flow chart and block diagram in accompanying drawing, it is illustrated that according to the system of the various embodiments of the application,
Architectural framework in the cards, function and the operation of method and computer program product.This point
On, each square frame in flow chart or block diagram can represent a module, program segment or code
A part, the part of the module, program segment or code is used for comprising one or more
The executable instruction of logic function as defined in realizing.It should also be noted that being used as replacement at some
In realization, the function of being marked in square frame can also be with different from the order marked in accompanying drawing hair
It is raw.For example, two square frames succeedingly represented can essentially be performed substantially in parallel, they
Sometimes it can also perform in the opposite order, this is depending on involved function.It is also noted that
It is, each square frame in block diagram and/or flow chart and the square frame in block diagram and/or flow chart
Combination, can be realized with the special hardware based system of defined function or operation is performed,
Or can be realized with the combination of specialized hardware and computer instruction.
Being described in module involved in the embodiment of the present application can be real by way of software
It is existing, it can also be realized by way of hardware.Described module can also be arranged on processing
In device, for example, can be described as:A kind of processor includes counseling problem receiving module, correlation
Spend acquisition module and answer language material sending module.Wherein, the title of these modules is in certain situation
Under do not constitute restriction to the module in itself, for example, counseling problem receiving module can also quilt
It is described as " module for receiving the counseling problem of user's input ".
As on the other hand, present invention also provides a kind of nonvolatile computer storage media,
The nonvolatile computer storage media can be described in above-described embodiment included in device
Nonvolatile computer storage media;Can also be individualism, without non-in supplying terminal
Volatile computer storage medium.Above-mentioned nonvolatile computer storage media be stored with one or
The multiple programs of person, when one or more of programs are performed by an equipment so that described
Equipment:The counseling problem of user's input is received, wherein, the counseling problem is made up of word;
The Question-Answering Model that the counseling problem is inputted into training in advance obtains the counseling problem and asked with first
The degree of correlation of each answer language material in corpus is answered, wherein, the Question-Answering Model is described for determining
The degree of correlation of counseling problem and answer language material;To there is maximum relation degree in the question and answer corpus
Answer language material be sent to the user.
Above description is only the preferred embodiment of the application and saying to institute's application technology principle
It is bright.It will be appreciated by those skilled in the art that invention scope involved in the application, is not limited
In the technical scheme of the particular combination of above-mentioned technical characteristic, do not departed from while should cover yet
In the case of the inventive concept, it is combined by above-mentioned technical characteristic or its equivalent feature
Formed by other technical schemes.Such as features described above and (but not limited to) disclosed herein
Technical characteristic with similar functions carries out technical scheme formed by replacement mutually.
Claims (12)
1. a kind of human-machine intelligence's answering method, it is characterised in that methods described includes:
The counseling problem of user's input is received, wherein, the counseling problem is made up of word;
The Question-Answering Model that the counseling problem is inputted into training in advance obtains the counseling problem and the
The degree of correlation of each answer language material in one question and answer corpus, wherein, the Question-Answering Model is used to determine
The degree of correlation of the counseling problem and answer language material;
Answer language material with maximum relation degree in the first question and answer corpus is sent to described
User.
2. according to the method described in claim 1, it is characterised in that described by the consulting
The Question-Answering Model of problem input training in advance is obtained in the counseling problem and the first question and answer corpus
Before the degree of correlation step of each answer language material, methods described also includes:
The second question and answer corpus is obtained, wherein, the second question and answer corpus includes multiple question and answer
Language material pair, the problem of each question and answer language material is to by corresponding to language material and answer language material are constituted;
Based on the counseling problem in the second question and answer corpus each problem language material it is related
Degree, to each question and answer language material in the second question and answer corpus to being ranked up;
According to the sequence, the problem of choosing predetermined number language material, and by described problem language material and
Corresponding answer language material constitutes the first question and answer corpus.
3. method according to claim 2, it is characterised in that methods described also includes:
The step of setting up Question-Answering Model, including:
Obtain the second question and answer corpus;
Extract the semantic feature of each question and answer language material pair and non-semantic spy in the second question and answer corpus
Levy, determine the similarity of the semantic feature of the question and answer language material centering problem language material and answer language material
And the similarity of the non-semantic feature of described problem language material and answer language material;
The similarity of similarity and the non-semantic feature based on the semantic feature, is trained
To the Question-Answering Model.
4. method according to claim 3, it is characterised in that methods described also includes:
Each problem language material and corresponding answer language material in second corpus are obtained, with
Machine generates new question and answer language material pair, and by the new question and answer language material to the 3rd question and answer language material of composition
Storehouse.
5. method according to claim 4, it is characterised in that the extraction described second
The semantic feature and non-semantic feature of each question and answer language material pair, determine the question and answer in question and answer corpus
The similarity and described problem language material of the semantic feature of language material centering problem language material and answer language material
With the similarity of the non-semantic feature of answer language material, including:
Determine the semantic feature similarity and non-semantic feature similarity the step of, including:
The first problem language material and the first answer language material of the first question and answer language material pair are obtained, wherein, institute
The first question and answer language material is stated to in the second question and answer corpus or the 3rd question and answer corpus
Any question and answer language material pair;
Determine the first problem language material and the Longest Common Substring in the first answer language material
Length value is used as the first similarity;
Determine the literal similarity of the first problem language material and the first answer language material as
Two similarities;
Determine the N-gram of the first problem language material and the first answer language material matching degree
As third phase like degree;
Determine the term vector of the first problem language material and the term vector of the first answer language material
Angle is used as the 4th similarity;
Determine the master of the first answer language material described in the theme feature vector sum of the first problem language material
The similarity of topic characteristic vector is used as the 5th similarity;
Wherein, first similarity, second similarity and the third phase are institute like degree
State the similarity of the non-semantic feature of first problem language material and the first answer language material, described the
Four similarities and the 5th similarity are the first problem language material and the first answer language material
Semantic feature similarity;
The step of similarity based on the similarity for determining the semantic feature and non-semantic feature
Suddenly, each first question and answer language material in the second question and answer corpus and the 3rd question and answer corpus is determined
First similarity of centering first problem language material and the first answer language material, the second similarity, the 3rd
Similarity, the 4th similarity and the 5th similarity.
6. method according to claim 5, it is characterised in that the determination described first
First answer language material described in the theme feature vector sum of problem language material theme feature vector it is similar
Spend as the 5th similarity, including:
According to each question and answer language in the second question and answer corpus and/or the 3rd question and answer corpus
Material to the problem of the corresponding theme of the language material and corresponding theme of answer language material, generate subject analysis mould
Type, wherein, the subject analysis model be used for by any one problem language material or any one answer
Case language material is converted into the theme feature vector of described problem language material or the answer language material;
The first problem and first answer input subject analysis model are obtained into corresponding institute
State the theme feature of the first answer language material described in the theme feature vector sum of first problem language material to
Amount;
By the theme of the first answer language material described in the theme feature vector sum of the first problem language material
The similarity of characteristic vector is used as the 5th similarity.
7. a kind of human-machine intelligence's question and answer system, it is characterised in that described device includes:
Counseling problem receiving module, is configured to receive the counseling problem of user's input, wherein,
The counseling problem is made up of word;
Degree of correlation acquisition module, is configured to input the counseling problem question and answer of training in advance
Model obtains the degree of correlation of each answer language material in the counseling problem and the first question and answer corpus, its
In, the Question-Answering Model is used for the degree of correlation for determining the counseling problem and answer language material;
Answer language material sending module, is configured to will have maximum in the first question and answer corpus
The answer language material of the degree of correlation is sent to the user.
8. device according to claim 7, it is characterised in that described device also includes:
First question and answer corpus comprising modules, are configured to obtain the second question and answer corpus, wherein,
The second question and answer corpus includes multiple question and answer language materials pair, and each question and answer language material is to by correspondence
The problem of language material and answer language material composition;
Based on the counseling problem in the second question and answer corpus each problem language material it is related
Degree, to each question and answer language material in the second question and answer corpus to being ranked up;
According to the sequence, the problem of choosing predetermined number language material, and by described problem language material and
Corresponding answer language material constitutes the first question and answer corpus.
9. device according to claim 8, peculiar to be, described device also includes:
Question-Answering Model sets up module, is configured to set up Question-Answering Model, including:
Second question and answer corpus acquiring unit, is configured to obtain the second question and answer corpus;
Similarity determining unit, is configured to extract each question and answer language in the second question and answer corpus
Expect to semantic feature and non-semantic feature, determine the question and answer language material centering problem language material and to answer
The similarity and described problem language material of the semantic feature of case language material and the non-semantic spy of answer language material
The similarity levied;
Question-Answering Model training unit, is configured to similarity based on the semantic feature and described
The similarity of non-semantic feature, training obtains the Question-Answering Model.
10. device according to claim 9, it is characterised in that described device also includes:
3rd question and answer corpus comprising modules, are configured to obtain each in second corpus
Problem language material and corresponding answer language material, generate new question and answer language material pair at random, and by institute
New question and answer language material is stated to the 3rd question and answer corpus of composition.
11. device according to claim 10, it is characterised in that the similarity is determined
Unit configuration specifically for:
Determine the semantic feature similarity and non-semantic feature similarity the step of, including:
The first problem language material and the first answer language material of the first question and answer language material pair are obtained, wherein, institute
The first question and answer language material is stated to in the second question and answer corpus or the 3rd question and answer corpus
Any question and answer language material pair;
Determine the first problem language material and the Longest Common Substring in the first answer language material
Length value is used as the first similarity;
Determine the literal similarity of the first problem language material and the first answer language material as
Two similarities;
Determine the N-gram of the first problem language material and the first answer language material matching degree
As third phase like degree;
Determine the term vector of the first problem language material and the term vector of the first answer language material
Angle is used as the 4th similarity;
Determine the master of the first answer language material described in the theme feature vector sum of the first problem language material
The similarity of topic characteristic vector is used as the 5th similarity;
Wherein, first similarity, second similarity and the third phase are institute like degree
State the similarity of the non-semantic feature of first problem language material and the first answer language material, described the
Four similarities and the 5th similarity are the first problem language material and the first answer language material
Semantic feature similarity;
The step of similarity based on the similarity for determining the semantic feature and non-semantic feature
Suddenly, each first question and answer language material in the second question and answer corpus and the 3rd question and answer corpus is determined
First similarity of centering first problem language material and the first answer language material, the second similarity, the 3rd
Similarity, the 4th similarity and the 5th similarity.
12. device according to claim 11, it is characterised in that the similarity is determined
Unit configuration is further used for:
According to each question and answer language in the second question and answer corpus and/or the 3rd question and answer corpus
Material to the problem of the corresponding theme of the language material and corresponding theme of answer language material, generate subject analysis mould
Type, wherein, the subject analysis model be used for by any one problem language material or any one answer
Case language material is converted into the theme feature vector of described problem language material or the answer language material;
The first problem and first answer input subject analysis model are obtained into corresponding institute
State the theme feature of the first answer language material described in the theme feature vector sum of first problem language material to
Amount;
By the theme of the first answer language material described in the theme feature vector sum of the first problem language material
The similarity of characteristic vector is used as the 5th similarity.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610262740.1A CN107305578A (en) | 2016-04-25 | 2016-04-25 | Human-machine intelligence's answering method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610262740.1A CN107305578A (en) | 2016-04-25 | 2016-04-25 | Human-machine intelligence's answering method and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107305578A true CN107305578A (en) | 2017-10-31 |
Family
ID=60150498
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610262740.1A Pending CN107305578A (en) | 2016-04-25 | 2016-04-25 | Human-machine intelligence's answering method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107305578A (en) |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108170853A (en) * | 2018-01-19 | 2018-06-15 | 广东惠禾科技发展有限公司 | A kind of chat language material method for self-cleaning, device and user terminal |
CN109033428A (en) * | 2018-08-10 | 2018-12-18 | 深圳市磐创网络科技有限公司 | A kind of intelligent customer service method and system |
CN109086400A (en) * | 2018-07-30 | 2018-12-25 | 深圳追科技有限公司 | The self checking method and device of customer service robot |
CN109388697A (en) * | 2018-10-10 | 2019-02-26 | 成都小多科技有限公司 | Handle the method, apparatus and electric terminal of question and answer corpus |
CN109670029A (en) * | 2018-12-28 | 2019-04-23 | 百度在线网络技术(北京)有限公司 | For determining the method, apparatus, computer equipment and storage medium of problem answers |
CN109857850A (en) * | 2019-01-18 | 2019-06-07 | 深圳壹账通智能科技有限公司 | Counsel requests processing method, device, computer equipment and storage medium |
CN110019715A (en) * | 2017-12-08 | 2019-07-16 | 阿里巴巴集团控股有限公司 | Response determines method, apparatus, equipment, medium and system |
CN110069613A (en) * | 2019-04-28 | 2019-07-30 | 河北省讯飞人工智能研究院 | A kind of reply acquisition methods and device |
CN110489533A (en) * | 2019-07-09 | 2019-11-22 | 深圳追一科技有限公司 | Interactive method and relevant device |
WO2020133360A1 (en) * | 2018-12-29 | 2020-07-02 | 深圳市优必选科技有限公司 | Question text matching method and apparatus, computer device and storage medium |
CN111831800A (en) * | 2019-08-13 | 2020-10-27 | 北京嘀嘀无限科技发展有限公司 | Question-answer interaction method, device, equipment and storage medium |
CN112418875A (en) * | 2020-10-21 | 2021-02-26 | 航天信息股份有限公司 | Cross-platform tax intelligent customer service corpus migration method and device |
CN113779223A (en) * | 2021-09-15 | 2021-12-10 | 广州网律互联网科技有限公司 | Artificial intelligence service method, system and equipment based on deep learning |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102637192A (en) * | 2012-02-17 | 2012-08-15 | 清华大学 | Method for answering with natural language |
CN103049433A (en) * | 2012-12-11 | 2013-04-17 | 微梦创科网络科技(中国)有限公司 | Automatic question answering method, automatic question answering system and method for constructing question answering case base |
CN103577557A (en) * | 2013-10-21 | 2014-02-12 | 北京奇虎科技有限公司 | Device and method for determining capturing frequency of network resource point |
CN103577556A (en) * | 2013-10-21 | 2014-02-12 | 北京奇虎科技有限公司 | Device and method for obtaining association degree of question and answer pair |
CN103810218A (en) * | 2012-11-14 | 2014-05-21 | 北京百度网讯科技有限公司 | Problem cluster-based automatic asking and answering method and device |
US20150066482A1 (en) * | 2009-10-19 | 2015-03-05 | Gil Fuchs | Sytem and method for use of semantic understanding in storage, searching, and providing of data or other content information |
CN104598445A (en) * | 2013-11-01 | 2015-05-06 | 腾讯科技(深圳)有限公司 | Automatic question-answering system and method |
CN104679910A (en) * | 2015-03-25 | 2015-06-03 | 北京智齿博创科技有限公司 | Intelligent answering method and system |
-
2016
- 2016-04-25 CN CN201610262740.1A patent/CN107305578A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20150066482A1 (en) * | 2009-10-19 | 2015-03-05 | Gil Fuchs | Sytem and method for use of semantic understanding in storage, searching, and providing of data or other content information |
CN102637192A (en) * | 2012-02-17 | 2012-08-15 | 清华大学 | Method for answering with natural language |
CN103810218A (en) * | 2012-11-14 | 2014-05-21 | 北京百度网讯科技有限公司 | Problem cluster-based automatic asking and answering method and device |
CN103049433A (en) * | 2012-12-11 | 2013-04-17 | 微梦创科网络科技(中国)有限公司 | Automatic question answering method, automatic question answering system and method for constructing question answering case base |
CN103577557A (en) * | 2013-10-21 | 2014-02-12 | 北京奇虎科技有限公司 | Device and method for determining capturing frequency of network resource point |
CN103577556A (en) * | 2013-10-21 | 2014-02-12 | 北京奇虎科技有限公司 | Device and method for obtaining association degree of question and answer pair |
CN104598445A (en) * | 2013-11-01 | 2015-05-06 | 腾讯科技(深圳)有限公司 | Automatic question-answering system and method |
CN104679910A (en) * | 2015-03-25 | 2015-06-03 | 北京智齿博创科技有限公司 | Intelligent answering method and system |
Non-Patent Citations (2)
Title |
---|
王一方: "客服机器人答案排序学习方法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
陈俊杰: "网络购物环境下的问句答案匹配方法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110019715A (en) * | 2017-12-08 | 2019-07-16 | 阿里巴巴集团控股有限公司 | Response determines method, apparatus, equipment, medium and system |
CN110019715B (en) * | 2017-12-08 | 2023-07-14 | 阿里巴巴集团控股有限公司 | Response determination method, device, equipment, medium and system |
CN108170853B (en) * | 2018-01-19 | 2020-06-19 | 广东惠禾科技发展有限公司 | Chat corpus self-cleaning method and device and user terminal |
CN108170853A (en) * | 2018-01-19 | 2018-06-15 | 广东惠禾科技发展有限公司 | A kind of chat language material method for self-cleaning, device and user terminal |
CN109086400A (en) * | 2018-07-30 | 2018-12-25 | 深圳追科技有限公司 | The self checking method and device of customer service robot |
CN109033428A (en) * | 2018-08-10 | 2018-12-18 | 深圳市磐创网络科技有限公司 | A kind of intelligent customer service method and system |
CN109033428B (en) * | 2018-08-10 | 2021-09-10 | 深圳市磐创网络科技有限公司 | Intelligent customer service method and system |
CN109388697A (en) * | 2018-10-10 | 2019-02-26 | 成都小多科技有限公司 | Handle the method, apparatus and electric terminal of question and answer corpus |
CN109670029A (en) * | 2018-12-28 | 2019-04-23 | 百度在线网络技术(北京)有限公司 | For determining the method, apparatus, computer equipment and storage medium of problem answers |
CN109670029B (en) * | 2018-12-28 | 2021-09-07 | 百度在线网络技术(北京)有限公司 | Method, apparatus, computer device and storage medium for determining answers to questions |
WO2020133360A1 (en) * | 2018-12-29 | 2020-07-02 | 深圳市优必选科技有限公司 | Question text matching method and apparatus, computer device and storage medium |
CN109857850A (en) * | 2019-01-18 | 2019-06-07 | 深圳壹账通智能科技有限公司 | Counsel requests processing method, device, computer equipment and storage medium |
CN110069613A (en) * | 2019-04-28 | 2019-07-30 | 河北省讯飞人工智能研究院 | A kind of reply acquisition methods and device |
CN110489533A (en) * | 2019-07-09 | 2019-11-22 | 深圳追一科技有限公司 | Interactive method and relevant device |
CN111831800A (en) * | 2019-08-13 | 2020-10-27 | 北京嘀嘀无限科技发展有限公司 | Question-answer interaction method, device, equipment and storage medium |
CN112418875A (en) * | 2020-10-21 | 2021-02-26 | 航天信息股份有限公司 | Cross-platform tax intelligent customer service corpus migration method and device |
CN112418875B (en) * | 2020-10-21 | 2024-03-26 | 航天信息股份有限公司 | Cross-platform tax intelligent customer service corpus migration method and device |
CN113779223A (en) * | 2021-09-15 | 2021-12-10 | 广州网律互联网科技有限公司 | Artificial intelligence service method, system and equipment based on deep learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107305578A (en) | Human-machine intelligence's answering method and device | |
CN110019701B (en) | Method for question answering service, question answering service system and storage medium | |
CN108287864B (en) | Interest group dividing method, device, medium and computing equipment | |
CN111353037B (en) | Topic generation method and device and computer readable storage medium | |
CN107220386A (en) | Information-pushing method and device | |
CN107305575A (en) | The punctuate recognition methods of human-machine intelligence's question answering system and device | |
CN107491547A (en) | Searching method and device based on artificial intelligence | |
CN105893344A (en) | User semantic sentiment analysis-based response method and device | |
CN111368548A (en) | Semantic recognition method and device, electronic equipment and computer-readable storage medium | |
CN107273348B (en) | Topic and emotion combined detection method and device for text | |
CN108121800A (en) | Information generating method and device based on artificial intelligence | |
CN107885785A (en) | Text emotion analysis method and device | |
CN111414561B (en) | Method and device for presenting information | |
CN107436916A (en) | The method and device of intelligent prompt answer | |
CN106484766A (en) | Searching method based on artificial intelligence and device | |
CN107862058A (en) | Method and apparatus for generating information | |
CN107783962A (en) | Method and device for query statement | |
CN112100353A (en) | Man-machine conversation method and system, computer device and medium | |
CN113627797A (en) | Image generation method and device for employee enrollment, computer equipment and storage medium | |
CN107291774A (en) | Error sample recognition methods and device | |
CN111666513A (en) | Page processing method and device, electronic equipment and readable storage medium | |
CN107256224A (en) | A kind of generation method of the key element structure of knowledge, searching method, apparatus and system | |
CN109190116A (en) | Semantic analytic method, system, electronic equipment and storage medium | |
CN116662495A (en) | Question-answering processing method, and method and device for training question-answering processing model | |
CN110704707A (en) | Service recommendation method and device based on user portrait |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171031 |