CN109739991A - The heterogeneous electric power data Uniform semantic theme modeling method of mode based on sharing feature space - Google Patents
The heterogeneous electric power data Uniform semantic theme modeling method of mode based on sharing feature space Download PDFInfo
- Publication number
- CN109739991A CN109739991A CN201811402024.4A CN201811402024A CN109739991A CN 109739991 A CN109739991 A CN 109739991A CN 201811402024 A CN201811402024 A CN 201811402024A CN 109739991 A CN109739991 A CN 109739991A
- Authority
- CN
- China
- Prior art keywords
- node
- feature space
- network
- heterogeneous
- similarity
- 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
- 238000000034 method Methods 0.000 title claims abstract description 35
- 239000011159 matrix material Substances 0.000 claims abstract description 28
- 230000009466 transformation Effects 0.000 claims abstract description 21
- 238000012545 processing Methods 0.000 claims description 4
- 241000208340 Araliaceae Species 0.000 claims description 3
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims description 3
- 235000003140 Panax quinquefolius Nutrition 0.000 claims description 3
- 230000002708 enhancing effect Effects 0.000 claims description 3
- 235000008434 ginseng Nutrition 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 2
- 238000005457 optimization Methods 0.000 claims description 2
- 238000004458 analytical method Methods 0.000 abstract description 6
- 230000006870 function Effects 0.000 description 21
- 238000011160 research Methods 0.000 description 6
- 230000008569 process Effects 0.000 description 4
- 238000007418 data mining Methods 0.000 description 3
- 241001269238 Data Species 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 101001059734 Thermococcus litoralis (strain ATCC 51850 / DSM 5473 / JCM 8560 / NS-C) Trehalose/maltose-binding protein MalE Proteins 0.000 description 1
- 238000009412 basement excavation Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of heterogeneous electric power data Uniform semantic theme modeling methods of mode based on sharing feature space, include the following steps, 1) analysis text data and the characteristics of image data, data prediction is carried out, corresponding character representation is obtained;2) according to the incidence relation between text data and image data, construct heterogeneous information network, by the link of heterogeneous network interior joint, the matrix of a linear transformation based on network structure is constructed, the node content of different modalities is transformed into identical feature space by the matrix of a linear transformation;3) similarity of the calculate node in sharing feature space and the similarity in primitive network, and loss function is constructed with this, obtain final Uniform semantic theme modeling result.Consider that the mode of data content is rich, structure semantics and the node content for comprehensively considering heterogeneous network are semantic, effectively solve the multi-modal data modeling problem in complex information network.
Description
Technical field
The invention belongs to power business technical field of data processing, and in particular to a kind of mode based on sharing feature space
Heterogeneous electric power data Uniform semantic theme modeling method.
Background technique
Heterogeneous information network HIN (Heterogeneous Information Networks) is made of multiclass node
It is internal it is interconnected can more between complete characterization complex data feature and data relationship network structure.Since 2009
After Sun et al. proposes the concept of HIN network and proposes meta-path Research Thinking, the research of heterogeneous information network becomes
One of research hotspot of the field of data mining.More and more researchers are the art teaches many brand-new data minings to grind
Study carefully method and task, typically there is similarity retrieval, information classification, information recommendation, link analysis.Similarity retrieval is heterogeneous letter
The basic problem in network structure is ceased, Sun et al. is put forward for the first time the PathSim method based on heterogeneous path to measure the same category
Relationship between node.Xiong et al. proposes PS-join (path-basedsimilarity join) method for returning
Top-k each other most like object pair.The excavation of text subject is also the pith in heterogeneous information network research, research
It focuses in the work that HIN method for digging is combined with topic model.Deng et al. proposes that TMBP model passes through deviation propagation pair
Topic model carries out canonical constraint in HIN, and further, they propose joint probability topic model while modeling heterogeneous information
The semanteme of multi-class data in network.In recent years, CHINC model introduces knowledge base and assists heterogeneous information net as indirect supervision information
Text cluster problem in network.
With RankClus, PathSim etc. is the heterogeneous information network modeling method of representative due to lacking to data point content
Semantic learning ability, therefore there are biggish limitations in the learning tasks of labyrinth data.Data primitive character
Contain more noise and redundancy, often in order to obtain the feature of stronger semantic meaning representation ability, in nineteen ninety, Deerwester
Et al. propose latent semantic analysis (Latent Semantic Analysis or LSA) method breakthroughly, completely with data
Semantic structure inside the mode learning data of driving.Hofmann in 1999 has redefined LSA from the angle of probability theory, mentions
Randomization latent semantic analysis model, i.e. PLSA (ProbabilisticLatentSemantic Analysis) are gone out.Into one
Step ground, Blei et al. model the generation of new document and its theme distribution in introducing Dirichlet prior distribution in 2003
Habit problem proposes LDA (Latent Dirichlet Allocate).Based on the two basic probability topics of PLSA and LDA
Model, it is subsequent and have many extended models: as Guo et al. constructs regularization factors based on the degree of association between inquiry by introducing
The study of topic is constrained, to guarantee that its topic distribution that learns of similarity high document is consistent;The HDP that Teh et al. is proposed
(HierarchicalDirichlet Process) model can go the number for learning topic automatically with non-parametric method.In addition, also
Auxiliary information is introduced into LDA to learn theme by many research work, this can also regard a kind of relationship of heterogeneous information as.
For example, typically there is Author-Topic model to model author field subject information while modeling text subject structure
ACT (Author-Conference-Topic) model modeling author's conferencing information etc..In addition to it is above-mentioned receive significant attention it is general
Outside rate method, the work of many non-randomization methods has also been emerged in large numbers in recent years.Non- randomization method is mainly by linear algebra
Tool to model topic.Wherein most work substitutes the SVD in original LSA with Non-negative Matrix Factorization (NMF) and decomposes.NMF
Former word-document matrix is resolved into two non-negative low-rank submatrixs, respectively indicates the relationship between word-topic and topic-document.
Particularly, it is limited Boltzmann machine (Restricted Boltzmann Machine, RBM) recently and is also applied to implicit theme
Modeling, relative to topic models such as PLSA, LDA, RBM has better efficiency in terms of theme deduction, and in file retrieval
It is achieved than LDA better performance in tasks with classification etc..
It is widely applied in field, if power business data are in the prevalence of interrelated and different types of data, such as
Heterogeneous information net is constituted including multi-modal datas, this kind of complex datas such as starting text, images.In practical problem, data are closed
All there is complicated heterogeneities for the supervision message of system, data content and data, since node has the feature of different modalities
It indicates, in the case where unprocessed, can not be modeled in the same model, this proposes the modeling method of heterogeneous information net
Stern challenge.
Summary of the invention
It is an object of the invention to overcome the deficiencies of the prior art and provide a kind of mode based on sharing feature space is heterogeneous
Electric power data Uniform semantic theme modeling method further considers that the mode of data content is rich, comprehensively considers heterogeneous network
Structure semantics and node content it is semantic, effectively solve the multi-modal data modeling problem in complex information network.
The present invention is achieved by the following technical solutions:
A kind of heterogeneous electric power data Uniform semantic theme modeling method of mode based on sharing feature space, including following step
Suddenly,
1) the characteristics of analyzing text data and image data, carries out data prediction, obtains corresponding character representation;
2) according to the incidence relation between text data and image data, heterogeneous information network is constructed, heterogeneous network is passed through
The link of interior joint constructs the matrix of a linear transformation based on network structure, by the matrix of a linear transformation by the node of different modalities
Content Transformation is into identical feature space;
3) similarity of the calculate node in sharing feature space and the similarity in primitive network, and constructed with this
Loss function obtains final Uniform semantic theme modeling result by minimizing loss function Optimal Parameters.
In the above-mentioned technical solutions, the data prediction in the step 1) includes:
, by text vector, text is obtained by segmenting, going stop words, word2vec to operate for the word content of text
This character representation;
The pixel content of image, by picture vectorization, is obtained by gray processing, geometric transformation, picture enhancing operation
The character representation of picture.
In the above-mentioned technical solutions, text, the characteristic dimension of graph node are different, by the matrix of a linear transformation by text section
Point and graph node are mapped to unified r3It ties up in the communal space.
In the above-mentioned technical solutions, the multi-modal heterogeneous information net of electric power is constructed according to the association of text data and image data
Network:
G<V,E>
Wherein G indicates that heterogeneous information network, V indicate set of network nodes, and E indicates network node link set, includes in V
Text node x and graph node z.
In the above-mentioned technical solutions, by the link of heterogeneous network interior joint, the matrix based on network structure is constructed:
Wherein, | V | indicate the node number in heterogeneous network, wijV is indicated for the element in L matrix, only 0 and 1,1iWith
vjBetween directly link, 0 indicate viAnd vjBetween do not directly link.
In the above-mentioned technical solutions, if the characteristic dimension of text and graph node is respectively r1And r2, using linear transformation square
Battle arrayWithText node x and graph node z are mapped to r3In dimension space, new text section is obtained
PointWith picture nodeIt is specific as follows:
In the above-mentioned technical solutions, the step of similarity of the calculate node in sharing feature space includes:
In r3It ties up in sharing feature space, the similarity of same type content node indicates are as follows:
In r3It ties up in sharing feature space, the similarity of same type and different types of content node indicates are as follows:
Wherein,It is jointly true by Λ and Π
Fixed.
In the above-mentioned technical solutions, the matrix of a linear transformation based on network structure based on construction, obtains node and arrives other
The link situation of all nodes indicates that concrete condition is as follows:
Wherein,Indicate node xiTo the link situation of node n,Indicate node xjTo the link situation of node n,Indicate node ziTo the link situation of node n,Indicate node zjTo the link situation of node n.Respectively indicate node xi、xj、zi、zjSituation is linked between all nodes;
In order to measure similarity of the node in primitive network, decision function d is constructed using Euclidean distance, it will be between node
Neighbor relationships are converted into real number representation, and the numerical values recited of decision function is enable to indicate similarity of the node in primitive network,
Similarity of the same type content node in primitive network can indicate are as follows:
Similarity of the different type content node in primitive network can indicate are as follows:
Wherein, d (xi,xj)、d(zi,zj) and d (xi,zj) indicate to establish decision using first routing information between node
Function.
In the above-mentioned technical solutions, loss function is constructed, specific as follows:
The loss function E of same type content nodesIt can indicate are as follows:
Wherein, dsIndicate similarity of the same type content node in primitive network, ssIt indicates in r3Tie up sharing feature space
The similarity of middle same type content node;
The loss function E of different type content nodedIt can indicate are as follows:
Wherein, ddIndicate similarity of the different type content node in primitive network, ssIt indicates in r3It is empty to tie up sharing feature
Between middle different type content node similarity.
In the above-mentioned technical solutions, parameter optimization is carried out based on loss function, obtains Optimal Parameters transformation matrixWith
It is to stop adjusting ginseng, output phase by judging whether the variation of loss function restrains or reach maximum number of iterations
Answer in r3Tie up text node new in sharing feature spaceWith picture nodeOtherwise continue parameter to update and optimize.
The advantages and benefits of the present invention are:
Semantic modeling is according to power business multi-modal data, the characteristics of analyzing text data and image data, carries out data
Pretreatment, obtains corresponding character representation.Secondly, being constructed heterogeneous according to the incidence relation between text data and image data
Information network and accordingly based on the matrix of network structure, is converted the node content of different modalities by the matrix of a linear transformation
Into identical feature space.Finally, similarity of the calculate node in sharing feature space and the phase in primitive network
Loss function is constructed like degree, and with this, by judging whether the variation of loss function restrains or reach maximum number of iterations, is obtained
Final Uniform semantic modeling result greatly promotes application of the heterogeneous information Web Mining in complex task, while being also pair
The important exploration and extension of machine Learning Theory have important theoretical value and broad application prospect.
Detailed description of the invention
Fig. 1 is a kind of heterogeneous electric power data Uniform semantic theme modeling method of mode based on sharing feature space of the invention
Flow chart.
It for those of ordinary skill in the art, without creative efforts, can be according to above attached
Figure obtains other relevant drawings.
Specific embodiment
In order to enable those skilled in the art to better understand the solution of the present invention, combined with specific embodiments below furtherly
Bright technical solution of the present invention.
The present invention mainly uses data mining theories and method to analyze heterogeneous information data, in order to guarantee system
It operates normally, in specific implementation, it is desirable that used computer platform is equipped with the memory for being not less than 8G, and CPU core calculation is not low
In 4 and the 64 bit manipulation systems of not low 2.6GHz, the Windows 7 of dominant frequency and the above version, and install oracle database,
The Kinds of Essential Software environment such as python3 version.
As shown in Figure 1, the heterogeneous electric power data Uniform semantic of a kind of mode based on sharing feature space provided by the invention
Theme modeling method executes the following steps in order:
Step 1) inputs multi-modal power business data, pre-processes to multi-modal data, obtains the spy of data
Sign indicates;
Step 2) is based on step 1) and constructs heterogeneous information network according to multi-modal power business data.
The link for passing through heterogeneous network interior joint simultaneously, constructs the matrix based on network structure.
Step 3) is based on the heterogeneous information network exported in step 1) and step 2), constructs the shared spy of mode heterogeneous nodes
Space learning is levied, wherein text, the characteristic dimension of graph node are different, by the matrix of a linear transformation by text node and image section
Point is mapped to unified r3It ties up in the communal space, similarity of the calculate node in sharing feature space and in primitive network
Similarity, and loss function is constructed with this, then again all types of nodes is carried out with unified semantic modeling.
In step 1): the power business data of input, specific as follows:
Step 1.1) pre-processes multi-modal power business data.
The word content of text, by text vector, is obtained by segmenting, going stop words, word2vec etc. to operate
The character representation of text;
The pixel content of image, by picture vectorization, is obtained by operations such as gray processing, geometric transformation, picture enhancings
To the character representation of picture.
Step 2.1) is based on the output of step 1.1), and it is multi-modal to construct electric power according to the association of text data and image data
Heterogeneous information network:
G<V,E>
Wherein G indicates that heterogeneous information network, V indicate set of network nodes, and E indicates network node link set, includes in V
Text node x and graph node z.
Step 2.2) is based on the output in step 2.1), and by the link of heterogeneous network interior joint, building is based on network knot
The matrix of structure:
Wherein, | V | indicate the node number in heterogeneous network, wijV is indicated for the element in L matrix, only 0 and 1,1iWith
vjBetween directly link, 0 indicate viAnd vjBetween do not directly link.
In step 3), the sharing feature space learning building of the mode heterogeneous nodes is as follows:
Step 3.1) assumes that the characteristic dimension of text and graph node is respectively r1And r2, using the matrix of a linear transformationWithText node x and graph node z are mapped to r3In dimension space, new text node is obtainedWith picture nodeIt is specific as follows:
Step 3.2) is based on the output of step 3.1), similarity of the calculate node in sharing feature space.
In order to reduce the information loss of text and image in mapping process, it is desirable that table of the node in sharing feature space
Show the neighbor relationships that can be maintained at as far as possible in primitive network.
In r3It ties up in sharing feature space, the similarity of same type content node can indicate are as follows:
In r3It ties up in sharing feature space, the similarity of same type and different types of content node can indicate are as follows:
Wherein,It is jointly true by Λ and Π
Fixed.
Step 3.3) obtains node to other all sections based on the matrix based on network structure constructed in step 2.2)
The link situation of point indicates that concrete condition is as follows:
Wherein,Indicate node xiTo the link situation of node n,Indicate node xjTo the link situation of node n,Indicate node ziTo the link situation of node n,Indicate node zjTo the link situation of node n.Respectively indicate node xi、xj、zi、zjSituation is linked between all nodes.Link situation is node
Association situation, that is, have link then for 1 without being then 0, or specific weight can be set according to specific correlation degree.
The output that step 3.4) is based on step 3.3) uses Euclidean to measure similarity of the node in primitive network
Distance building decision function d, converts real number representation for the neighbor relationships between node, enables the numerical values recited table of decision function
Show neighbour degree of the node in primitive network, i.e. similarity.
Similarity of the same type content node in primitive network can indicate are as follows:
Similarity of the different type content node in primitive network can indicate are as follows:
Wherein, d (xi,xj)、d(zi,zj) and d (xi,zj) indicate to establish decision using first routing information between node
Function.
Step 3.5) is based on the output of step 3.3) and step 3.4), constructs loss function, specific as follows:
The loss function E of same type content nodesIt can indicate are as follows:
Wherein, dsIndicate similarity of the same type content node in primitive network, ssIt indicates in r3Tie up sharing feature space
The similarity of middle same type content node.
The loss function E of different type content nodedIt can indicate are as follows:
Wherein, ddIndicate similarity of the different type content node in primitive network, ssIt indicates in r3It is empty to tie up sharing feature
Between middle different type content node similarity.
Step 3.6) is based on the loss function in step 3.5), Optimal Parameters transformation matrixWith
It is to stop adjusting ginseng, output phase by judging whether the variation of loss function restrains or reach maximum number of iterations
Answer in r3Tie up text node new in sharing feature spaceWith picture nodeOtherwise continue parameter to update and optimize;
Finally export new text nodeWith picture nodeIt ends here.
The spatially relative terms such as "upper", "lower", "left", "right" have been used in embodiment for ease of explanation, have been used for
Relationship of the elements or features relative to another elements or features shown in explanatory diagram.It should be understood that in addition to figure
Shown in except orientation, spatial terminology is intended to include the different direction of device in use or operation.For example, if in figure
Device be squeezed, the element for being stated as being located at other elements or feature "lower" will be located into other elements or feature "upper".
Therefore, exemplary term "lower" may include both upper and lower orientation.Device, which can be positioned in other ways, (to be rotated by 90 ° or position
In other orientation), it can be interpreted accordingly used herein of the opposite explanation in space.
Moreover, the relational terms of such as " first " and " second " or the like are used merely to one with another with identical
The component of title distinguishes, without necessarily requiring or implying between these components there are any this actual relationship or
Sequentially.
Illustrative description has been done to the present invention above, it should explanation, the case where not departing from core of the invention
Under, any simple deformation, modification or other skilled in the art can not spend the equivalent replacement of creative work equal
Fall into protection scope of the present invention.
Claims (10)
1. a kind of heterogeneous electric power data Uniform semantic theme modeling method of mode based on sharing feature space, it is characterised in that:
Include the following steps,
1) the characteristics of analyzing text data and image data, carries out data prediction, obtains corresponding character representation;
2) according to the incidence relation between text data and image data, heterogeneous information network is constructed, by saving in heterogeneous network
The link of point constructs the matrix of a linear transformation based on network structure, by the matrix of a linear transformation by the node content of different modalities
It is transformed into identical feature space;
3) similarity of the calculate node in sharing feature space and the similarity in primitive network, and loss is constructed with this
Function obtains final Uniform semantic theme modeling result by minimizing loss function Optimal Parameters.
2. the heterogeneous electric power data Uniform semantic theme of a kind of mode based on sharing feature space according to claim 1 is built
Mould method, it is characterised in that: the data prediction in the step 1) includes:
, by text vector, text is obtained by segmenting, going stop words, word2vec to operate for the word content of text
Character representation;
, by picture vectorization, picture is obtained by gray processing, geometric transformation, picture enhancing operation for the pixel content of image
Character representation.
3. the heterogeneous electric power data Uniform semantic theme of a kind of mode based on sharing feature space according to claim 1 is built
Mould method, it is characterised in that: text, the characteristic dimension of graph node are different, by the matrix of a linear transformation by text node and figure
As node is mapped to unified r3It ties up in the communal space.
4. the heterogeneous electric power data Uniform semantic theme of a kind of mode based on sharing feature space according to claim 1 is built
Mould method, it is characterised in that: the multi-modal heterogeneous information network of electric power is constructed according to the association of text data and image data:
G<V,E>
Wherein G indicates that heterogeneous information network, V indicate set of network nodes, and it includes text in V that E, which indicates network node link set,
Node x and graph node z.
5. the heterogeneous electric power data Uniform semantic theme of a kind of mode based on sharing feature space according to claim 4 is built
Mould method, it is characterised in that: by the link of heterogeneous network interior joint, construct the matrix based on network structure:
Wherein, | V | indicate the node number in heterogeneous network, wijV is indicated for the element in L matrix, only 0 and 1,1iAnd vjIt
Between directly link, 0 indicate viAnd vjBetween do not directly link.
6. the heterogeneous electric power data Uniform semantic theme of a kind of mode based on sharing feature space according to claim 5 is built
Mould method, it is characterised in that: the characteristic dimension for setting text and graph node is respectively r1And r2, using the matrix of a linear transformationWithText node x and graph node z are mapped to r3In dimension space, new text node is obtained
With picture nodeIt is specific as follows:
7. the heterogeneous electric power data Uniform semantic theme of a kind of mode based on sharing feature space according to claim 5 is built
Mould method, it is characterised in that: the step of similarity of the calculate node in sharing feature space includes:
In r3It ties up in sharing feature space, the similarity of same type content node indicates are as follows:
In r3It ties up in sharing feature space, the similarity of same type and different types of content node indicates are as follows:
Wherein,It is to be determined jointly by Λ and Π.
8. the heterogeneous electric power data Uniform semantic theme of a kind of mode based on sharing feature space according to claim 5 is built
Mould method, it is characterised in that:
The matrix of a linear transformation based on network structure based on construction, obtain node to other all nodes link situation table
Show, concrete condition is as follows:
Wherein,Indicate node xiTo the link situation of node n,Indicate node xjTo the link situation of node n,
Indicate node ziTo the link situation of node n,Indicate node zjTo the link situation of node n.Point
It Biao Shi not node xi、xj、zi、zjSituation is linked between all nodes;
In order to measure similarity of the node in primitive network, decision function d is constructed using Euclidean distance, by the neighbour between node
Transformation is real number representation, and the numerical values recited of decision function is enable to indicate similarity of the node in primitive network,
Similarity of the same type content node in primitive network can indicate are as follows:
Similarity of the different type content node in primitive network can indicate are as follows:
Wherein, d (xi,xj)、d(zi,zj) and d (xi,zj) indicate to establish decision function using first routing information between node.
9. the heterogeneous electric power data Uniform semantic theme of a kind of mode based on sharing feature space according to claim 5 is built
Mould method, it is characterised in that: building loss function, specific as follows:
The loss function E of same type content nodesIt can indicate are as follows:
Wherein, dsIndicate similarity of the same type content node in primitive network, ssIt indicates in r3It ties up same in sharing feature space
The similarity of type content node;
The loss function E of different type content nodedIt can indicate are as follows:
Wherein, ddIndicate similarity of the different type content node in primitive network, ssIt indicates in r3It ties up in sharing feature space
The similarity of different type content node.
10. the heterogeneous electric power data Uniform semantic theme of a kind of mode based on sharing feature space according to claim 9
Modeling method, it is characterised in that: parameter optimization is carried out based on loss function, obtains Optimal Parameters transformation matrixWith
It is to stop adjusting ginseng, output is corresponding by judging whether the variation of loss function restrains or reach maximum number of iterations
In r3Tie up text node new in sharing feature spaceWith picture nodeOtherwise continue parameter to update and optimize.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811402024.4A CN109739991A (en) | 2018-11-22 | 2018-11-22 | The heterogeneous electric power data Uniform semantic theme modeling method of mode based on sharing feature space |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811402024.4A CN109739991A (en) | 2018-11-22 | 2018-11-22 | The heterogeneous electric power data Uniform semantic theme modeling method of mode based on sharing feature space |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109739991A true CN109739991A (en) | 2019-05-10 |
Family
ID=66358136
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811402024.4A Pending CN109739991A (en) | 2018-11-22 | 2018-11-22 | The heterogeneous electric power data Uniform semantic theme modeling method of mode based on sharing feature space |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109739991A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110826635A (en) * | 2019-11-12 | 2020-02-21 | 曲阜师范大学 | Sample clustering and feature identification method based on integration non-negative matrix factorization |
CN113268858A (en) * | 2021-04-25 | 2021-08-17 | 贵州电网有限责任公司 | Low-voltage distribution network equipment data modeling method combining non-negative matrix factorization |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105205096A (en) * | 2015-08-18 | 2015-12-30 | 天津中科智能识别产业技术研究院有限公司 | Text modal and image modal crossing type data retrieval method |
CN105893612A (en) * | 2016-04-26 | 2016-08-24 | 中国科学院信息工程研究所 | Consistency expression method for multi-source heterogeneous big data |
CN106844518A (en) * | 2016-12-29 | 2017-06-13 | 天津中科智能识别产业技术研究院有限公司 | A kind of imperfect cross-module state search method based on sub-space learning |
CN108132968A (en) * | 2017-12-01 | 2018-06-08 | 西安交通大学 | Network text is associated with the Weakly supervised learning method of Semantic unit with image |
US20180293505A1 (en) * | 2017-04-06 | 2018-10-11 | Universite Paris Descartes | Method for clustering nodes of a textual network taking into account textual content, computer-readable storage device and system implementing said method |
-
2018
- 2018-11-22 CN CN201811402024.4A patent/CN109739991A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105205096A (en) * | 2015-08-18 | 2015-12-30 | 天津中科智能识别产业技术研究院有限公司 | Text modal and image modal crossing type data retrieval method |
CN105893612A (en) * | 2016-04-26 | 2016-08-24 | 中国科学院信息工程研究所 | Consistency expression method for multi-source heterogeneous big data |
CN106844518A (en) * | 2016-12-29 | 2017-06-13 | 天津中科智能识别产业技术研究院有限公司 | A kind of imperfect cross-module state search method based on sub-space learning |
US20180293505A1 (en) * | 2017-04-06 | 2018-10-11 | Universite Paris Descartes | Method for clustering nodes of a textual network taking into account textual content, computer-readable storage device and system implementing said method |
CN108132968A (en) * | 2017-12-01 | 2018-06-08 | 西安交通大学 | Network text is associated with the Weakly supervised learning method of Semantic unit with image |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110826635A (en) * | 2019-11-12 | 2020-02-21 | 曲阜师范大学 | Sample clustering and feature identification method based on integration non-negative matrix factorization |
CN110826635B (en) * | 2019-11-12 | 2023-04-18 | 曲阜师范大学 | Sample clustering and feature identification method based on integration non-negative matrix factorization |
CN113268858A (en) * | 2021-04-25 | 2021-08-17 | 贵州电网有限责任公司 | Low-voltage distribution network equipment data modeling method combining non-negative matrix factorization |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112529168B (en) | GCN-based attribute multilayer network representation learning method | |
Soibelman et al. | Management and analysis of unstructured construction data types | |
CN111523051A (en) | Social interest recommendation method and system based on graph volume matrix decomposition | |
Peng et al. | Dynamic network embedding via incremental skip-gram with negative sampling | |
CN114048331A (en) | Knowledge graph recommendation method and system based on improved KGAT model | |
CN112613602A (en) | Recommendation method and system based on knowledge-aware hypergraph neural network | |
Fang et al. | Transfer learning across networks for collective classification | |
CN111737535A (en) | Network characterization learning method based on element structure and graph neural network | |
WO2019154411A1 (en) | Word vector retrofitting method and device | |
CN107679035A (en) | A kind of information intent detection method, device, equipment and storage medium | |
CN109739991A (en) | The heterogeneous electric power data Uniform semantic theme modeling method of mode based on sharing feature space | |
Zhang et al. | Deep learning based link prediction with social pattern and external attribute knowledge in bibliographic networks | |
Sun et al. | Graph force learning | |
Huang et al. | Intangible cultural heritage management using machine learning model: a case study of Northwest Folk Song Huaer | |
KR20210121921A (en) | Method and device for extracting key keywords based on keyword joint appearance network | |
Hu et al. | MBRep: Motif-based representation learning in heterogeneous networks | |
Yang et al. | A novel evolutionary method to search interesting association rules by keywords | |
CN117669046A (en) | Data integration method and system for intelligent decision of automobile assembly process | |
Liu et al. | Scientific Paper Classification Based on Graph Neural Network with Hypergraph Self-attention Mechanism | |
Zhang et al. | A subgraph sampling method for training large-scale graph convolutional network | |
Cheng et al. | BHONEM: Binary high-order network embedding methods for networked-guarantee loans | |
Liao et al. | An Automatic and Unified Consistency Verification Rule and Method of SG-CIM Model | |
Liu et al. | TD-GAT: Graph neural network for fault diagnosis knowledge graph | |
Cao et al. | Energy-aware design: predicting building performance from layout graphs | |
Dai | Design and research of big data technology based on e-commerce platform |
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: 20190510 |