CN104572820B - The generation method and device of model, importance acquisition methods and device - Google Patents
The generation method and device of model, importance acquisition methods and device Download PDFInfo
- Publication number
- CN104572820B CN104572820B CN201410723276.2A CN201410723276A CN104572820B CN 104572820 B CN104572820 B CN 104572820B CN 201410723276 A CN201410723276 A CN 201410723276A CN 104572820 B CN104572820 B CN 104572820B
- Authority
- CN
- China
- Prior art keywords
- entry
- accuracy rate
- output valve
- candidate family
- normalized
- 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.)
- Active
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
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiments of the invention provide a kind of generation method of model and device, importance acquisition methods and device.On the one hand, for the embodiment of the present invention by least one in the discrimination accuracy rate between the sequence accuracy rate between the importance accuracy rate according to the entry obtained, entry and entry, M candidate family of structure, M is the integer more than 0；So as to using the M candidate family, obtain the individual normalized candidate family output valves of M of the entry；And then the M normalized candidate family output valves are assessed using assessment models, to obtain object module output valve, using the candidate family corresponding to the object module output valve as object module.Therefore, technical scheme provided in an embodiment of the present invention can solve the problems, such as that the Reliability comparotive of the model for the importance information for obtaining entry in the prior art is low.
Description
【Technical field】
The present invention relates to Computer Applied Technology field, more particularly to a kind of generation method of model and device, importance
Acquisition methods and device.
【Background technology】
For given text, the importance of wherein each entry is calculated exactly, and then can apply to follow-up
Search or semantic analysis etc..For example, under search scene, during user input query text, some words are included in the query text
Bar, wherein there can be redundancy entry, if scanned for query text really, search efficiency can be influenceed and reduce search knot
The quality of fruit.Therefore, it is necessary to carry out importance calculating to the entry in query text, then using wherein importance it is higher one
A little entries go to scan for, and remove redundancy entry therein.
In the prior art, having can be according to the model of given entry output importance or importance sorting, these mould
Type can ensure the accuracy of the numerical value of the importance of the entry of output, or can ensure two words for belonging to same text
The accuracy of importance sorting between bar.If however, both needing to obtain the numerical value of the importance of entry, while also need to word
The importance sorting of bar, then current model can not all meet, in addition, other letters of the importance of entry can not be obtained
Cease, discrimination, the span of importance between the importance of such as entry, therefore, currently used for obtaining the important of entry
The Reliability comparotive for spending the model of information is low.
【The content of the invention】
In view of this, the embodiments of the invention provide a kind of generation method of model and device, importance acquisition methods and
Device, can solve the problems, such as that the Reliability comparotive of the model for the importance information for obtaining entry in the prior art is low.
The one side of the embodiment of the present invention, there is provided a kind of generation method of model, including：
According to the importance accuracy rate of entry obtained, the sequence accuracy rate between entry and the discrimination standard between entry
It is at least one in true rate, M candidate family is built, M is the integer more than 0；
Using the M candidate family, M normalized candidate family output valves of the entry are obtained；
The M normalized candidate family output valves are assessed using assessment models, it is defeated to obtain object module
Go out value, using the candidate family corresponding to the object module output valve as object module.
Aspect as described above and any possible implementation, it is further provided a kind of implementation, the foundation obtain
The importance accuracy rate of entry, at least one in the sequence accuracy rate between entry and the discrimination accuracy rate between entry
Individual, before building M candidate family, methods described also includes：
Using initial model, the initial model output valve of the entry is obtained；
According to the initial model output valve of the entry and the entry other entries in the text initial model
Output valve, obtain the normalized initial model output valve of the entry；
According to the normalized initial model output valve of the entry, the importance accuracy rate of the entry is obtained；With/
Or, according to the normalized initial model output valve of the entry or the initial model output valve of the entry, described in acquisition
Sequence accuracy rate between entry；And/or the normalized initial model output valve according to the entry, obtain the entry
Between discrimination accuracy rate.
Aspect as described above and any possible implementation, it is further provided a kind of implementation, the entry
The number of importance accuracy rate is N number of, and the number of the sequence accuracy rate between the entry is P, the discrimination between entry
The number of accuracy rate is Q, and N, P and Q are positive integer, and is asynchronously 0, and at least one in N, P and Q is more than or equal to
2, the discrimination between the sequence accuracy rate and entry according between the importance accuracy rate of entry obtained, entry is accurate
It is at least one in rate, M candidate family is built, including：
The area between sequence accuracy rate between importance accuracy rate, the entry and the entry according to the entry
It is at least one in indexing accuracy rate, K target accuracy rate is obtained, K is more than 1 and is less than or equal toInteger；
According to the K target accuracy rate, the first model parameter of initial model is adjusted, to obtain M second
Model parameter；
According to the M the second model parameters, the M candidate family is built.
Aspect as described above and any possible implementation, it is further provided a kind of implementation, it is described according to institute
K target accuracy rate is stated, the first model parameter of initial model is adjusted, to obtain M the second model parameters, including：
According to the K target accuracy rate and default accuracy rate threshold value, acquisition is more than or equal to the accuracy rate
M target accuracy rate of threshold value；
Derivative operation is carried out respectively to each target accuracy rate in the M target accuracy rate, to obtain M ladder
Angle value；
According to each Grad in the M Grad, the first model parameter of the initial model is entered respectively
Row adjustment, to obtain M the second model parameters.
Aspect as described above and any possible implementation, it is further provided a kind of implementation, it is described to utilize institute
M candidate family is stated, obtains M normalized candidate family output valves of the entry, including：
Using the M candidate family, M candidate family output valve of the entry is obtained；
Each candidate family output valve, the entry according to the entry other entries in the text candidate family
Output valve, obtain M normalized candidate family output valves of the entry.
Aspect as described above and any possible implementation, it is further provided a kind of implementation, the assessment mould
Type includes the standard output value of entry institute all entries in the text, it is described using assessment models to described M normalization
Candidate family output valve assessed, to obtain object module output valve, including：
For each candidate family, K word of normalized candidate family output valve highest in the text is determined
Bar；
R entry of standard output value highest described in the text is determined, R is the integer more than K；
According to standard described in K entry of normalized candidate family output valve highest in the text, the text
R entry of output valve highest, obtain the assessment result of each normalized candidate family output valve；
According to the assessment result, a normalized time is selected from the M normalized candidate family output valves
Modeling type output valve, to be used as the object module output valve.
The one side of the embodiment of the present invention, there is provided a kind of importance acquisition methods, including：
Obtain pending text；
Cutting word processing is carried out to the pending text, to obtain at least one entry；
Using object module, the object module output valve of each entry in acquisition at least one entry, to make
For the importance of each entry；
Wherein, the object module is to be generated using the generation method of model described above.
The one side of the embodiment of the present invention, there is provided a kind of generating means of model, including：
Construction unit, for the sequence accuracy rate and entry between the importance accuracy rate according to the entry obtained, entry
Between discrimination accuracy rate in it is at least one, build M candidate family, M is integer more than 0；
First acquisition unit, for utilizing the M candidate family, obtain M normalized candidate's moulds of the entry
Type output valve；
Assessment unit, for being assessed using assessment models the M normalized candidate family output valves, to obtain
Object module output valve is obtained, using the candidate family corresponding to the object module output valve as object module.
Aspect as described above and any possible implementation, it is further provided a kind of implementation, described device is also
Including：
Second acquisition unit, for using initial model, obtaining the initial model output valve of the entry；And it is used for
According to the initial model output valve of the entry and the entry other entries in the text initial model output valve, obtain
Obtain the normalized initial model output valve of the entry；It is and defeated for the normalized initial model according to the entry
Go out value, obtain the importance accuracy rate of the entry；And/or normalized initial model output valve according to the entry or
The initial model output valve of entry described in person, obtain the sequence accuracy rate between the entry；And/or according to the entry
Normalized initial model output valve, obtain the discrimination accuracy rate between the entry.
Aspect as described above and any possible implementation, it is further provided a kind of implementation, the entry
The number of importance accuracy rate is N number of, and the number of the sequence accuracy rate between the entry is P, the discrimination between entry
The number of accuracy rate is Q, and N, P and Q are positive integer, and is asynchronously 0, and at least one in N, P and Q is more than or equal to
2, the construction unit, it is specifically used for：
The area between sequence accuracy rate between importance accuracy rate, the entry and the entry according to the entry
It is at least one in indexing accuracy rate, K target accuracy rate is obtained, K is more than 1 and is less than or equal toInteger；
According to the K target accuracy rate, the first model parameter of initial model is adjusted, to obtain M second
Model parameter；
According to the M the second model parameters, the M candidate family is built.
Aspect as described above and any possible implementation, it is further provided a kind of implementation, the structure are single
Member is used to, according to the K target accuracy rate, be adjusted the first model parameter of initial model, to obtain M the second moulds
During shape parameter, it is specifically used for：
According to the K target accuracy rate and default accuracy rate threshold value, acquisition is more than or equal to the accuracy rate
M target accuracy rate of threshold value；
Derivative operation is carried out respectively to each target accuracy rate in the M target accuracy rate, to obtain M ladder
Angle value；
According to each Grad in the M Grad, the first model parameter of the initial model is entered respectively
Row adjustment, to obtain M the second model parameters.
Aspect as described above and any possible implementation, it is further provided a kind of implementation, described first obtains
Unit is taken, is specifically used for：
Using the M candidate family, M candidate family output valve of the entry is obtained；
Each candidate family output valve, the entry according to the entry other entries in the text candidate family
Output valve, obtain M normalized candidate family output valves of the entry.
Aspect as described above and any possible implementation, it is further provided a kind of implementation, the assessment mould
Type include the entry all entries in the text standard output value, the assessment unit, be specifically used for：
For each candidate family, K word of normalized candidate family output valve highest in the text is determined
Bar；
R entry of standard output value highest described in the text is determined, R is the integer more than K；
According to standard described in K entry of normalized candidate family output valve highest in the text, the text
R entry of output valve highest, obtain the assessment result of each normalized candidate family output valve；
According to the assessment result, a normalized time is selected from the M normalized candidate family output valves
Modeling type output valve, to be used as the object module output valve.
The one side of the embodiment of the present invention, there is provided a kind of importance acquisition device, including：
Acquiring unit, for obtaining pending text；
Cutting word unit, for carrying out cutting word processing to the pending text, to obtain at least one entry；
Processing unit, for using object module, obtaining the target mould of each entry at least one entry
Type output valve, using the importance as each entry；
Wherein, the object module is to be generated using the generating means of model described above.
As can be seen from the above technical solutions, the embodiment of the present invention has the advantages that：
The embodiment of the present invention is by merging multiple importance targets of entry come the structure of implementation model, and then by mould
The output valve of type is assessed to determine an optimal models, and in the prior art, model is merely able to ensure the entry of output
The accuracy of the numerical value of importance, or it is merely able to the importance sorting that guarantee belongs between two entries of same text
The technical scheme of accuracy is compared, and the constructed model with determination of the embodiment of the present invention, can meet multiple importances simultaneously
Correlated condition, ensure the accuracy of the related data of entry, the area between sequence or entry such as the numerical value, entry of importance
Indexing, so as to solve the problems, such as that the Reliability comparotive of the model for the importance information for obtaining entry in the prior art is low, is carried
The reliability of high model, and improve the accuracy of output data.
【Brief description of the drawings】
In order to illustrate the technical solution of the embodiments of the present invention more clearly, below by embodiment it is required use it is attached
Figure is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the present invention, for this area
For those of ordinary skill, without having to pay creative labor, it can also be obtained according to these accompanying drawings other attached
Figure.
Fig. 1 is the schematic flow sheet of the generation method for the model that the embodiment of the present invention is provided；
Fig. 2 is the schematic flow sheet of the acquisition methods for the importance that the embodiment of the present invention is provided；
Fig. 3 is the functional block diagram of the generating means of the model provided by the embodiment of the present invention；
Fig. 4 is the functional block diagram of the acquisition device for the importance that the embodiment of the present invention is provided.
【Embodiment】
In order to be better understood from technical scheme, the embodiment of the present invention is retouched in detail below in conjunction with the accompanying drawings
State.
It will be appreciated that described embodiment is only part of the embodiment of the present invention, rather than whole embodiments.Base
Embodiment in the present invention, those of ordinary skill in the art obtained under the premise of creative work is not made it is all its
Its embodiment, belongs to the scope of protection of the invention.
The term used in embodiments of the present invention is only merely for the purpose of description specific embodiment, and is not intended to be limiting
The present invention." one kind ", " described " and "the" of singulative used in the embodiment of the present invention and appended claims
It is also intended to including most forms, unless context clearly shows that other implications.
It should be appreciated that term "and/or" used herein is only a kind of incidence relation for describing affiliated partner, represent
There may be three kinds of relations, for example, A and/or B, can be represented：Individualism A, while A and B be present, individualism B these three
Situation.In addition, character "/" herein, it is a kind of relation of "or" to typically represent forward-backward correlation object.
It will be appreciated that though in embodiments of the present invention may using term first, second etc. come descriptive model parameter, but
These keywords should not necessarily be limited by these terms.These terms are only used for model parameter being distinguished from each other out.For example, this is not being departed from
In the case of inventive embodiments scope, the first model parameter can also be referred to as the second model parameter, similarly, the second model ginseng
Number can also be referred to as the first model parameter.
Depending on linguistic context, word as used in this " if " can be construed to " ... when " or " when ...
When " or " in response to determining " or " in response to detection ".Similarly, depending on linguistic context, phrase " if it is determined that " or " if detection
(condition or event of statement) " can be construed to " when it is determined that when " or " in response to determine " or " when the detection (condition of statement
Or event) when " or " in response to detecting (condition or event of statement) ".
Embodiment one
The embodiment of the present invention provides a kind of generation method of model, refer to Fig. 1, it is provided by the embodiment of the present invention
The schematic flow sheet of the generation method of model, as illustrated, this method comprises the following steps：
S101, according to the importance accuracy rate of entry obtained, the sequence accuracy rate between entry and the area between entry
It is at least one in indexing accuracy rate, M candidate family is built, M is the integer more than 0.
S102, using the M candidate family, obtain M normalized candidate family output valves of the entry.
S103, the M normalized candidate family output valves are assessed using assessment models, to obtain target mould
Type output valve, using the candidate family corresponding to the object module output valve as object module.
Based on the generation method of above-mentioned model, before the embodiment of the present invention is to S101, methods described can also include step：
Obtain at least one in the discrimination accuracy rate between sequence accuracy rate and the entry between the importance accuracy rate of entry, entry
It is individual.
The step can specifically include：
First, using default initial model, the initial model output valve of the entry is obtained.Then, according to institute's predicate
The initial model output valve of bar and the entry other entries in the text initial model output valve, obtain the entry
Normalized initial model output valve.Finally, the normalized initial model output valve according to the entry, institute's predicate is obtained
The importance accuracy rate of bar；And/or normalized initial model output valve according to the entry or the entry is initial
Model output valve, obtain the sequence accuracy rate between the entry；And/or the normalized initial model according to the entry
Output valve, obtain the discrimination accuracy rate between the entry.
For example, for entry xij, using initial model, the following initial model output valve of the entry can be obtained：
Wherein, characteristic vector xijRepresent j-th of entry in i-th text.
Wherein,Represent the initial model output valve of j-th of entry in i-th text.
Wherein, αkRepresent the weighted value of kth decision tree.
Wherein, the initial model can be Gradient Iteration sorting tree (Gradient Boosted RankingTree,
GBRank) model, the GBRank models can be made up of many decision trees, and every decision tree can be according to input xijObtain
One output valve hk(xij), added up again after then the output valve of every decision tree is multiplied with the weighted value of decision tree, it is possible to
Obtain initial model output valve
Wherein, the output valve h of kth decision treek(xij) by the decision tree, the model parameter of itself determines, the model parameter
The disruptive features of decision tree, the characteristic value of decision tree and regressand value corresponding to leaf node etc. can be included but is not limited to.
For example, for the initial model output valve of entry, the initial model output valve can be carried out using equation below
Normalized, to obtain the normalized initial model output valve of entry：
Wherein,For the normalized initial model output valve of j-th of entry in i-th text.
Wherein,Represent the initial model output valve of j-th of entry in i-th text.
It should be noted that in i-th text is obtained the normalized initial model output valve of j-th of entry process
In, it is also desirable to using the entry other entries in the text initial model output valve, therefore can be obtained using above-mentioned formula
Obtain the initial model output valve of other entries.
For example, the normalized initial model output valve according to entryOr the initial model output valve of the entryUsing but be not limited in equation below any one importance accuracy rate for obtaining entry：
In above-mentioned formula,The importance accuracy rate of j-th of entry in i-th text is represented, for realizing to mould
Type optimizes so that model disclosure satisfy that the accuracy condition of the numerical value of the importance of the entry of output.niRepresent the i-th provision
The total number of entry in this；yijRepresent the default master pattern output valve of j-th of entry in i-th text；Represent to utilize
The normalized initial model output valve of j-th of entry in i-th text that the above method obtains.
For example, the normalized initial model output valve according to entryOr initial model output valveUsing but not
It is limited to the sequence accuracy rate between any one described entry of acquisition in equation below：
In above-mentioned formula,Represent the sequence of the importance in i-th text between j-th of entry and k-th of entry
Accuracy rate, model is optimized for realizing so that model disclosure satisfy that the sequence of the importance between the entry of output is accurate
True property condition.niRepresent the total number of entry in i-th text；Represent jth in i-th text using above method acquisition
The normalized initial model output valve of individual entry, similarly, k-th of word in i-th text can also be obtained using the above method
The normalized initial model output valve of barThe initial model output valve of j-th of entry in i-th text is represented,
Represent the initial model output valve of k-th of entry in i-th text.τ represents default preset parameter.
For example, the normalized initial model output valve according to entryUsing but be not limited to any one in equation below
The individual discrimination accuracy rate obtained between the entry：
In above-mentioned formula,Represent the differentiation of the importance in i-th text between j-th of entry and k-th of entry
Accuracy rate is spent, model is optimized for realizing so that model disclosure satisfy that the differentiation of the importance between the entry of output
The accuracy condition of degree.niRepresent the total number of entry in i-th text；Represent the i-th provision obtained using the above method
Normalized initial model output valve of j-th of entry, similarly, can also be obtained in i-th text using the above method in this
The normalized initial model output valve of k-th of entryyijRepresent the default master die of j-th of entry in i-th text
Type output valve, yikRepresent the default master pattern output valve of k-th of entry in i-th text.Represent the i-th provision
This all entriesThe variance that value is obtained, Var (yi*) represent i-th text all entries yi*The variance that value is obtained.
Based on the generation method of above-mentioned model, S101 method is specifically described the embodiment of the present invention.The step has
Body can include：
The number of formula that importance accuracy rate by obtaining entry is utilized can be N number of (N in such as above-mentioned example
Can be individual (P is equal to 6 in such as above-mentioned example) for P in the number for 5), obtaining the formula that the sequence accuracy rate between entry is utilized,
The number for obtaining the formula that the discrimination accuracy rate between entry is utilized can be Q (Q is equal to 6 in such as above-mentioned example), N, P
Be positive integer with Q, and N, P be 0 during Q differences, and it is at least one in N, P and Q be more than or equal to 2, so, accordingly,
The number for obtaining the importance accuracy rate of entry can be N number of (such as N is equal to 5), obtain the number of the sequence accuracy rate between entry
Mesh can be P (such as P is equal to 6), and the number for obtaining the discrimination accuracy rate between entry can be Q (such as Q is equal to 6).
It is possible, firstly, to sequence accuracy rate and institute's predicate between importance accuracy rate, the entry according to the entry
It is at least one in discrimination accuracy rate between bar, K target accuracy rate is obtained, K is more than 1 and is less than or equal to
Integer.Then, according to the K target accuracy rate, the first model parameter of initial model is adjusted, to obtain M
Second model parameter.Finally, according to the M the second model parameters, the M candidate family is built.
For example, sequence accuracy rate and the entry between importance accuracy rate, the entry according to the entry it
Between discrimination accuracy rate, utilize equation below obtain target accuracy rate：
In above-mentioned formula,Represent target accuracy rate.
Wherein, αpoint、αpair、αlistFor default three preset parameters, these three preset parameters are 1 with value.
It is understood that because the number of the importance accuracy rate of entry can be N number of, the sequence between entry is accurate
The number of rate can be P, and the number of the discrimination accuracy rate between entry can be Q, so the importance according to entry
The discrimination accuracy rate between sequence accuracy rate and the entry between accuracy rate, the entry, and permutation and combination principle,
K target accuracy rate can be obtained, K value may be greater than 1 and be less than or equal toInteger.
For example, in the embodiment of the present invention, according to K target accuracy rate, the first model parameter of initial model is entered
Row adjustment, it can be included but is not limited to obtaining the method for M the second model parameters：
First, according to the K target accuracy rate and default accuracy rate threshold value, each target accuracy rate is judged
Whether default accuracy rate threshold value is more than or equal to, to obtain in K target accuracy rate, more than or equal to the accuracy rate
M target accuracy rate of threshold value, K are more than M.Then, each target accuracy rate in the M target accuracy rate is distinguished
Derivative operation is carried out, to obtain M Grad.Finally, according to each Grad in the M Grad, respectively to institute
The first model parameter for stating initial model is adjusted, to obtain M the second model parameters.Conversely, for accurate less than described
The target accuracy rate of rate threshold value, represent that the adjustment of the first model parameter of the initial model need not be carried out, can be by initially
Model is directly as one of M candidate family.
For example, derivative operation can be carried out to target accuracy rate using equation below, to obtain Grad：
In above-mentioned formula,Represent target accuracy rate；Represent the initial model output of j-th of entry in i-th text
Value；αpoint、αpair、αlistFor default three preset parameters, these three preset parameters are 1 with value.Represent i-th
The importance accuracy rate of j-th of entry in text,Represent in i-th text between j-th of entry and k-th of entry
Sort accuracy rate,Represent the discrimination accuracy rate between j-th of entry and k-th of entry in i-th text.
It should be noted that for the second different model parameters, different candidate families can be obtained.For example, can be with
According to each second model parameter, model training is carried out using GBRank algorithms, structure includes the candidate family of some decision trees,
Therefore, using M the second model parameters, M candidate family can be constructed.Wherein, the second model parameter can include but not
It is limited to following parameter：Regressand value corresponding to leaf node in the disruptive features of decision tree, the characteristic value of decision tree and decision tree.
Based on the generation method of above-mentioned model, S102 method is specifically described the embodiment of the present invention.The step has
Body can include：
First, using the M candidate family, M candidate family output valve of the entry is obtained.Then, according to institute
Each candidate family output valve of predicate bar, the entry other entries in the text candidate family output valve, obtain institute
M normalized candidate family output valves of predicate bar.
For example, for entry xij, using initial model, following candidate family output valve can be obtained：
Wherein, characteristic vector xijRepresent j-th of entry in i-th text.
In this step,Represent the candidate family output valve of j-th of entry in i-th text.
In this step, αkRepresent the weighted value of kth decision tree in certain candidate family.
Wherein, the candidate family can be GBRank models, and the GBRank models can be made up of many decision trees,
Every decision tree can be according to input xijObtain an output valve hk(xij), then by the output valve of every decision tree and decision-making
The weighted value of tree adds up again after being multiplied, it is possible to obtains candidate family output valve
For example, for the candidate family output valve of entry, the candidate family output valve can be carried out using equation below
Normalized, to obtain the normalized candidate family output valve of entry：
Wherein,For the normalized candidate family output valve of j-th of entry in i-th text.
Wherein,Represent the candidate family output valve of j-th of entry in i-th text.
It is understood that in i-th text is obtained the normalized candidate family output valve of j-th of entry process
In, it is necessary to using the entry other entries in the text candidate family output valve.
It is understood that for each candidate family in M candidate family, the acquisition of the two formula may be by
Corresponding normalized candidate family output valve, it is hereby achieved that M normalized candidate family output valves.
It should be noted that initial model output valve and calculating of the above-mentioned two formula with calculating entry using initial model
The formula of normalized initial model output valve is identical, that is to say, that and the method that the output valve of model is obtained using model is identical,
The model only used is different, so the numerical value of model output is also different.
Based on the generation method of above-mentioned model, S103 method is specifically described the embodiment of the present invention.The step has
Body can include：
In the embodiment of the present invention, the standard that the assessment models can include but is not limited to each entry in each text is defeated
Go out value.
For example, M normalized candidate family output valves are assessed using the assessment models, to obtain target
The method of model output valve can include but is not limited to：
First, for each candidate family, K word of normalized candidate family output valve highest in the text is determined
Bar；And standard output value R entry of highest described in the text is determined, R is the integer more than K.Then, according to described in
K entry of normalized candidate family output valve highest in text, standard output value R word of highest described in the text
Bar, obtain the assessment result of each normalized candidate family output valve.Finally, according to the assessment result, from the M
A normalized candidate family output valve is selected in individual normalized candidate family output valve, using defeated as the object module
Go out value, can be using the candidate family corresponding to the object module output valve as target mould it is determined that after object module output valve
Type.
For example, being directed to each candidate family, normalized 2 entries of candidate family output valve highest in text are determined,
It is then determined that text Plays output valve 2 entries of highest.Then, normalized candidate corresponding to each candidate family is calculated
The hit rate of 2 entry hit criteria output valve 2 entries of highest of model output valve highest, and calculate each candidate's mould
The variance of 2 entries of normalized candidate family output valve highest corresponding to type, hit rate and/or variance are tied as assessment
Fruit.Finally, according to assessment result, an optimal normalized candidate is selected in M normalized candidate family output valves
Model output valve, the candidate family for exporting the optimal normalized candidate family output valve is exactly optimal models, so as to
As the object module in the embodiment of the present invention.
It is understood that in the embodiment of the present invention, pass through N number of importance accuracy rate to entry, P sequence accuracy rate
And the various combination of Q discrimination accuracy rate, all combinations can be enumerated, difference is constructed according to different combined results
Candidate family, then using the candidate family constructed as test sample, entry is carried out respectively using these candidate families
Calculate, to obtain candidate family output valve.Candidate family output valve is realized to candidate family finally by using assessment models
Assessment, to obtain optimal candidate family so that hereafter using the candidate family obtain entry importance information when, can
Discrimination between ranking results or entry is obtained between the importance numerical value of entry, entry, and then can be improved according to entry
Importance information acquisition search result when, the accuracy of search result, or, improve according to entry importance information carry out
During semantic analysis, the accuracy of semantic analysis result.
It should be noted that model of the prior art is for the entry in given text, otherwise it can only ensure to export
Entry between importance sequence accuracy, or can only ensure output entry importance accuracy, for example,
Entry a, the entry b of model output and entry c importance are respectively 1,0 and -1, or the importance of some entries of output
It is ordered as entry a=1>Entry b=0>Entry c=-1.However, the importance that many search systems require to obtain at present must expire
The importance of each entry of foot is non-negative, and importance is 1 with value, or, such as sometimes need entry
Importance meets some requirements, if desired for the numerical value using importance in 0.2, under these scenes, mould of the prior art
Type cannot reach these requirements.In technical scheme provided in an embodiment of the present invention, the output valve of model is normalized
Final output value of the normalized output valve obtained after processing as model, therefore the numerical value of the importance of entry can be ensured
For just, and the numerical value of the importance of entry is between 0 and 1, to meet the demand.
Embodiment two
The embodiment of the present invention provides a kind of acquisition methods of importance, refer to Fig. 2, it is provided by the embodiment of the present invention
Importance acquisition methods schematic flow sheet, as illustrated, this method comprises the following steps：
S201, obtain pending text.
S202, cutting word processing is carried out to the pending text, to obtain at least one entry.
S203, using object module, the object module output valve of each entry in acquisition at least one entry,
Using the importance as each entry；Wherein, the object module is to be generated using the generation method of above-mentioned model.
For example, the method for obtaining pending text can be the text for receiving user's input, such as query word (Query)；Or
Person, the voice messaging of user's input can also be received, according to text corresponding to voice messaging acquisition.
Preferably, cutting word processing can be carried out to pending text using dictionary for word segmentation, to obtain in the pending text
Comprising at least one entry.
Preferably, after at least one entry is obtained, generated using each entry as the generation method of above-mentioned model
Object module input, to cause object module to obtain an object module output valve, the object module to each entry
Output valve can serve as the importance of corresponding entry.
For example, using object module, the importance of the entry of acquisition can apply to realize the search of the text of input, such as
The particial entry higher according to importance scans for, rather than is scanned for according to whole entries, so as to improve search
Efficiency, and improve the quality and accuracy rate of search result.Or may be used also using object module, the importance of the entry of acquisition
So that applied to semantic analysis is realized, particial entry such as higher according to importance is analyzed the semanteme of user, to parse
The intention of user, and then corresponding operation is performed according to semantic analysis result, the embodiment of the present invention is not particularly limited to this.
The embodiment of the present invention further provides the device embodiment for realizing each step and method in above method embodiment.
Embodiment three
Fig. 3 is refer to, the functional block diagram of the generating means of its model provided by the embodiment of the present invention.As schemed
Show, the device includes：
Construction unit 301, for the sequence accuracy rate and word between the importance accuracy rate according to the entry obtained, entry
It is at least one in discrimination accuracy rate between bar, M candidate family is built, M is the integer more than 0；
First acquisition unit 302, for utilizing the M candidate family, obtain M normalized candidates of the entry
Model output valve；
Assessment unit 303, for being assessed using assessment models the M normalized candidate family output valves,
To obtain object module output valve, using the candidate family corresponding to the object module output valve as object module.
Optionally, described device also includes：
Second acquisition unit 304, for using initial model, obtaining the initial model output valve of the entry；And use
In the initial model output valve according to the entry and the entry other entries in the text initial model output valve,
Obtain the normalized initial model output valve of the entry；And for the normalized initial model according to the entry
Output valve, obtain the importance accuracy rate of the entry；And/or the normalized initial model output valve according to the entry
Or the initial model output valve of the entry, obtain the sequence accuracy rate between the entry；And/or according to the entry
Normalized initial model output valve, obtain the discrimination accuracy rate between the entry.
Preferably, the number of the importance accuracy rate of the entry is N number of, the number of the sequence accuracy rate between the entry
Mesh is P, and the number of the discrimination accuracy rate between entry is Q, and N, P and Q are positive integer, and is asynchronously 0, and N, P
It is more than or equal to 2 with least one in Q, the construction unit 301, is specifically used for：
The area between sequence accuracy rate between importance accuracy rate, the entry and the entry according to the entry
It is at least one in indexing accuracy rate, K target accuracy rate is obtained, K is more than 1 and is less than or equal toInteger；
According to the K target accuracy rate, the first model parameter of initial model is adjusted, to obtain M second
Model parameter；
According to the M the second model parameters, the M candidate family is built.
Preferably, the construction unit 301 is used for according to the K target accuracy rate, to the first model of initial model
Parameter is adjusted, and during obtaining M the second model parameters, is specifically used for：
According to the K target accuracy rate and default accuracy rate threshold value, acquisition is more than or equal to the accuracy rate
M target accuracy rate of threshold value；
Derivative operation is carried out respectively to each target accuracy rate in the M target accuracy rate, to obtain M ladder
Angle value；
According to each Grad in the M Grad, the first model parameter of the initial model is entered respectively
Row adjustment, to obtain M the second model parameters.
Preferably, the first acquisition unit 302, is specifically used for：
Using the M candidate family, M candidate family output valve of the entry is obtained；
Each candidate family output valve, the entry according to the entry other entries in the text candidate family
Output valve, obtain M normalized candidate family output valves of the entry.
Preferably, the assessment models include the entry all entries in the text standard output value, institute's commentary
Estimate unit 303, be specifically used for：
For each candidate family, K word of normalized candidate family output valve highest in the text is determined
Bar；
R entry of standard output value highest described in the text is determined, R is the integer more than K；
According to standard described in K entry of normalized candidate family output valve highest in the text, the text
R entry of output valve highest, obtain the assessment result of each normalized candidate family output valve；
According to the assessment result, a normalized time is selected from the M normalized candidate family output valves
Modeling type output valve, to be used as the object module output valve.
Because each unit in the present embodiment is able to carry out the method shown in Fig. 1, the part that the present embodiment is not described in detail,
Refer to the related description to Fig. 1.
Example IV
Fig. 4 is refer to, the functional block diagram of the acquisition device of its importance provided by the embodiment of the present invention.As schemed
Show, the device includes：
Acquiring unit 401, for obtaining pending text；
Cutting word unit 402, for carrying out cutting word processing to the pending text, to obtain at least one entry；
Processing unit 403, for using object module, obtaining the target of each entry at least one entry
Model output valve, using the importance as each entry；
Wherein, the object module is the generating means generation using the model described in above-mentioned Fig. 3.
Because each unit in the present embodiment is able to carry out the method shown in Fig. 2, the part that the present embodiment is not described in detail,
Refer to the related description to Fig. 2.
The technical scheme of the embodiment of the present invention has the advantages that：
The embodiment of the present invention passes through the sequence accuracy rate and word between the importance accuracy rate according to the entry obtained, entry
It is at least one in discrimination accuracy rate between bar, M candidate family is built, M is the integer more than 0；So as to utilize the M
Individual candidate family, obtain M normalized candidate family output valves of the entry；And then using assessment models to the M
Normalized candidate family output valve is assessed, to obtain object module output valve, object module output valve institute is right
The candidate family answered is as object module.Therefore, the embodiment of the present invention is realized by merging multiple importance targets of entry
The structure of model, an and then optimal models is determined by being assessed the output valve of model, and in the prior art, model
It is merely able to ensure the accuracy of the numerical value of the importance of the entry of output, or is merely able to ensure two that belong to same text
The technical scheme of the accuracy of importance sorting between entry is compared, the constructed model with determination of the embodiment of the present invention, energy
Enough correlated conditions for meeting multiple importances simultaneously, ensure the accuracy of the related data of entry, numerical value, entry such as importance
Sequence or entry between discrimination, so as to solve the model for the importance information for obtaining entry in the prior art
The problem of Reliability comparotive is low, the reliability of model is improved, and improve the accuracy of output data.
Further, since consider to merge multiple targets simultaneously, for single target therein, the complexity of problem reduces.
For example, in a text, the importance order of three entries, i.e. entry a, entry b and entry c is a>b>C, if importance
Order learnt it is accurate, then the numerical value of b importance is inevitable between a and c, therefore, accurate to the numerical value of importance
This true target optimizes space and greatly reduced, and reduces optimization complexity so that the importance of entry learns more accurate.
And consider to merge multiple targets simultaneously so that learning process convergence is very fast, and avoids overfitting to a certain extent.It is above-mentioned
Advantage, all individually consider that some target can not be brought.
It is apparent to those skilled in the art that for convenience and simplicity of description, the system of foregoing description,
The specific work process of device and unit, the corresponding process in preceding method embodiment is may be referred to, will not be repeated here.
In several embodiments provided by the present invention, it should be understood that disclosed system, apparatus and method can be with
Realize by another way.For example, device embodiment described above is only schematical, for example, the unit
Division, only a kind of division of logic function, can there is other dividing mode, for example, multiple units or group when actually realizing
Part can combine or be desirably integrated into another system, or some features can be ignored, or not perform.It is another, it is shown
Or the mutual coupling discussed or direct-coupling or communication connection can be by some interfaces, device or unit it is indirect
Coupling or communication connection, can be electrical, mechanical or other forms.
The unit illustrated as separating component can be or may not be physically separate, show as unit
The part shown can be or may not be physical location, you can with positioned at a place, or can also be distributed to multiple
On NE.Some or all of unit therein can be selected to realize the mesh of this embodiment scheme according to the actual needs
's.
In addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, can also
That unit is individually physically present, can also two or more units it is integrated in a unit.Above-mentioned integrated list
Member can both be realized in the form of hardware, can also be realized in the form of hardware adds SFU software functional unit.
The above-mentioned integrated unit realized in the form of SFU software functional unit, can be stored in one and computer-readable deposit
In storage media.Above-mentioned SFU software functional unit is stored in a storage medium, including some instructions are causing a computer
It is each that device (can be personal computer, server, or network equipment etc.) or processor (Processor) perform the present invention
The part steps of embodiment methods described.And foregoing storage medium includes：USB flash disk, mobile hard disk, read-only storage (Read-
Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disc or CD etc. it is various
Can be with the medium of store program codes.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all essences in the present invention
God any modification, equivalent substitution and improvements done etc., should be included within the scope of protection of the invention with principle.
Claims (12)
1. a kind of generation method of model, it is characterised in that methods described includes：
According to the importance accuracy rate of entry obtained, the sequence accuracy rate between entry and the discrimination accuracy rate between entry
In it is at least one, build M candidate family, M is integer more than 0；
Using the M candidate family, M normalized candidate family output valves of the entry are obtained；
The M normalized candidate family output valves are assessed using assessment models, to obtain object module output valve,
Using the candidate family corresponding to the object module output valve as object module；
It is described to utilize the M candidate family, M normalized candidate family output valves of the entry are obtained, including：
Using the M candidate family, M candidate family output valve of the entry is obtained；
Each candidate family output valve, the entry according to the entry other entries in the text candidate family output
Value, obtain M normalized candidate family output valves of the entry.
2. according to the method for claim 1, it is characterised in that the importance accuracy rate of the entry according to acquisition, word
It is at least one in the discrimination accuracy rate between sequence accuracy rate and entry between bar, it is described before building M candidate family
Method also includes：
Using initial model, the initial model output valve of the entry is obtained；
According to the initial model output valve of the entry and the entry in the text other entries initial model output
Value, obtain the normalized initial model output valve of the entry；
According to the normalized initial model output valve of the entry, the importance accuracy rate of the entry is obtained；And/or according to
According to the normalized initial model output valve of the entry or the initial model output valve of the entry, obtain the entry it
Between sequence accuracy rate；And/or the normalized initial model output valve according to the entry, obtain between the entry
Discrimination accuracy rate.
3. method according to claim 1 or 2, it is characterised in that the number of the importance accuracy rate of the entry is N
Individual, the number of the sequence accuracy rate between the entry is P, and the number of the discrimination accuracy rate between entry is Q, N, P
Be 0 or positive integer with Q, and be asynchronously 0, and it is in N, P and Q at least one be more than or equal to 2, it is described according to obtaining
At least one, structure in the discrimination accuracy rate between sequence accuracy rate and entry between the importance accuracy rate of entry, entry
M candidate family is built, including：
The discrimination between sequence accuracy rate between importance accuracy rate, the entry and the entry according to the entry
It is at least one in accuracy rate, K target accuracy rate is obtained, K is more than 1 and is less than or equal toInteger；
According to the K target accuracy rate, the first model parameter of initial model is adjusted, to obtain M the second models
Parameter；
According to the M the second model parameters, the M candidate family is built.
4. according to the method for claim 3, it is characterised in that it is described according to the K target accuracy rate, to initial model
The first model parameter be adjusted, to obtain M the second model parameters, including：
According to the K target accuracy rate and default accuracy rate threshold value, acquisition is more than or equal to the accuracy rate threshold value
M target accuracy rate；
Derivative operation is carried out respectively to each target accuracy rate in the M target accuracy rate, to obtain M gradient
Value；
According to each Grad in the M Grad, the first model parameter of the initial model is adjusted respectively
It is whole, to obtain M the second model parameters.
5. according to the method for claim 1, it is characterised in that the assessment models include institute in the text of entry institute
There is the standard output value of entry, it is described that the M normalized candidate family output valves are assessed using assessment models, with
Object module output valve is obtained, including：
For each candidate family, K entry of normalized candidate family output valve highest in the text is determined；
R entry of standard output value highest described in the text is determined, R is the integer more than K；
According to standard output described in K entry of normalized candidate family output valve highest in the text, the text
It is worth R entry of highest, obtains the assessment result of each normalized candidate family output valve；Knot is assessed according to described
Fruit, a normalized candidate family output valve is selected from the M normalized candidate family output valves, using as described
Object module output valve.
6. a kind of acquisition methods of importance, it is characterised in that methods described includes：
Obtain pending text；
Cutting word processing is carried out to the pending text, to obtain at least one entry；
Using object module, the object module output valve of each entry at least one entry is obtained, using as every
The importance of the individual entry；
Wherein, the object module is to be generated using the generation method of the model described in any claim in claim 1 to 5
's.
7. a kind of generating means of model, it is characterised in that described device includes：
Construction unit, between the sequence accuracy rate between the importance accuracy rate according to the entry obtained, entry and entry
Discrimination accuracy rate in it is at least one, build M candidate family, M is integer more than 0；
First acquisition unit, for utilizing the M candidate family, M normalized candidate families for obtaining the entry are defeated
Go out value；
Assessment unit, for being assessed using assessment models the M normalized candidate family output valves, to obtain mesh
Model output valve is marked, using the candidate family corresponding to the object module output valve as object module；
The first acquisition unit, is specifically used for：
Using the M candidate family, M candidate family output valve of the entry is obtained；
Each candidate family output valve, the entry according to the entry other entries in the text candidate family output
Value, obtain M normalized candidate family output valves of the entry.
8. device according to claim 7, it is characterised in that described device also includes：
Second acquisition unit, for using initial model, obtaining the initial model output valve of the entry；And for foundation
The initial model output valve of the entry and the entry other entries in the text initial model output valve, obtain institute
The normalized initial model output valve of predicate bar；And for the normalized initial model output valve according to the entry,
Obtain the importance accuracy rate of the entry；And/or the normalized initial model output valve or described according to the entry
The initial model output valve of entry, obtain the sequence accuracy rate between the entry；And/or the normalization according to the entry
Initial model output valve, obtain the discrimination accuracy rate between the entry.
9. the device according to claim 7 or 8, it is characterised in that the number of the importance accuracy rate of the entry is N
Individual, the number of the sequence accuracy rate between the entry is P, and the number of the discrimination accuracy rate between entry is Q, N, P
Be 0 or positive integer with Q, and be asynchronously 0, and it is at least one in N, P and Q be more than or equal to 2, the construction unit,
It is specifically used for：
The discrimination between sequence accuracy rate between importance accuracy rate, the entry and the entry according to the entry
It is at least one in accuracy rate, K target accuracy rate is obtained, K is more than 1 and is less than or equal toInteger；
According to the K target accuracy rate, the first model parameter of initial model is adjusted, to obtain M the second models
Parameter；
According to the M the second model parameters, the M candidate family is built.
10. device according to claim 9, it is characterised in that the construction unit is used for accurate according to the K target
Rate, the first model parameter of initial model is adjusted, during obtaining M the second model parameters, be specifically used for：
According to the K target accuracy rate and default accuracy rate threshold value, acquisition is more than or equal to the accuracy rate threshold value
M target accuracy rate；
Derivative operation is carried out respectively to each target accuracy rate in the M target accuracy rate, to obtain M gradient
Value；
According to each Grad in the M Grad, the first model parameter of the initial model is adjusted respectively
It is whole, to obtain M the second model parameters.
11. device according to claim 7, it is characterised in that the assessment models include entry institute in the text
The standard output value of all entries, the assessment unit, is specifically used for：
For each candidate family, K entry of normalized candidate family output valve highest in the text is determined；
R entry of standard output value highest described in the text is determined, R is the integer more than K；
According to standard output described in K entry of normalized candidate family output valve highest in the text, the text
It is worth R entry of highest, obtains the assessment result of each normalized candidate family output valve；
According to the assessment result, normalized candidate's mould is selected from the M normalized candidate family output valves
Type output valve, to be used as the object module output valve.
12. a kind of acquisition device of importance, it is characterised in that described device includes：
Acquiring unit, for obtaining pending text；
Cutting word unit, for carrying out cutting word processing to the pending text, to obtain at least one entry；
Processing unit, for using object module, the object module for obtaining each entry at least one entry to be defeated
Go out value, using the importance as each entry；
Wherein, the object module is to be given birth to using the generating means of the model described in any claim in claim 7 to 11
Into.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410723276.2A CN104572820B (en) | 2014-12-03 | 2014-12-03 | The generation method and device of model, importance acquisition methods and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410723276.2A CN104572820B (en) | 2014-12-03 | 2014-12-03 | The generation method and device of model, importance acquisition methods and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104572820A CN104572820A (en) | 2015-04-29 |
| CN104572820B true CN104572820B (en) | 2017-11-24 |
Family
ID=53088882
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410723276.2A Active CN104572820B (en) | 2014-12-03 | 2014-12-03 | The generation method and device of model, importance acquisition methods and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104572820B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107515876B (en) * | 2016-06-16 | 2020-12-04 | 阿里巴巴集团控股有限公司 | Feature model generation and application method and device |
| CN106503231B (en) * | 2016-10-31 | 2020-02-04 | 北京百度网讯科技有限公司 | Search method and device based on artificial intelligence |
| CN109919931B (en) * | 2019-03-08 | 2020-12-25 | 数坤（北京）网络科技有限公司 | Coronary stenosis degree evaluation model training method and evaluation system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102043843A (en) * | 2010-12-08 | 2011-05-04 | 百度在线网络技术（北京）有限公司 | Method and obtaining device for obtaining target entry based on target application |
| CN102789451A (en) * | 2011-05-16 | 2012-11-21 | 北京百度网讯科技有限公司 | Individualized machine translation system, method and translation model training method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120005566A1 (en) * | 2010-06-30 | 2012-01-05 | International Business Machines Corporation | Adding a comprehension marker to a social network text entry |
-
2014
- 2014-12-03 CN CN201410723276.2A patent/CN104572820B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102043843A (en) * | 2010-12-08 | 2011-05-04 | 百度在线网络技术（北京）有限公司 | Method and obtaining device for obtaining target entry based on target application |
| CN102789451A (en) * | 2011-05-16 | 2012-11-21 | 北京百度网讯科技有限公司 | Individualized machine translation system, method and translation model training method |
Non-Patent Citations (2)
| Title |
|---|
| Estimation and Evaluation of Conditional Asset Pricing Models;STEFAN NAGEL等;《The Journal of Finance》;20110523;第66卷(第3期);第873-909页 * |
| 基于支持向量机的聚类及文本分类研究;平源;《中国博士学位论文全文数据库信息科技辑》;20130115(第01期);第1-130页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104572820A (en) | 2015-04-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US9875294B2 (en) | Method and apparatus for classifying object based on social networking service, and storage medium | |
| US9401148B2 (en) | Speaker verification using neural networks | |
| CN109272988B (en) | Voice recognition method based on multi-path convolution neural network | |
| US10109272B2 (en) | Apparatus and method for training a neural network acoustic model, and speech recognition apparatus and method | |
| Ding et al. | Autospeech: Neural architecture search for speaker recognition | |
| CN112464638B (en) | Text clustering method based on improved spectral clustering algorithm | |
| CN106098059A (en) | customizable voice awakening method and system | |
| CN109918663A (en) | A kind of semantic matching method, device and storage medium | |
| CN108804677A (en) | In conjunction with the deep learning question classification method and system of multi-layer attention mechanism | |
| CN109857846B (en) | Method and device for matching user question and knowledge point | |
| CN111159363A (en) | Knowledge base-based question answer determination method and device | |
| CN109726387A (en) | Man-machine interaction method and system | |
| CN106060008B (en) | A kind of network intrusions method for detecting abnormality | |
| CN106096066A (en) | The Text Clustering Method embedded based on random neighbor | |
| CN112883741B (en) | Specific target emotion classification method based on dual-channel graph neural network | |
| Ochiai et al. | Automatic node selection for deep neural networks using group lasso regularization | |
| CN112115716A (en) | Service discovery method, system and equipment based on multi-dimensional word vector context matching | |
| CN104572820B (en) | The generation method and device of model, importance acquisition methods and device | |
| CN108509727B (en) | Model selection processing method and device in data modeling | |
| CN110085217A (en) | Phonetic navigation method, device and terminal device | |
| CN113326377A (en) | Name disambiguation method and system based on enterprise incidence relation | |
| CN113157919A (en) | Sentence text aspect level emotion classification method and system | |
| CN109558591A (en) | Chinese event detection method and device | |
| CN109410935A (en) | A kind of destination searching method and device based on speech recognition | |
| CN111079427A (en) | Junk mail identification method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |