CN105808689A - Drainage system entity semantic similarity measurement method based on artificial neural network - Google Patents
Drainage system entity semantic similarity measurement method based on artificial neural network Download PDFInfo
- Publication number
- CN105808689A CN105808689A CN201610119650.7A CN201610119650A CN105808689A CN 105808689 A CN105808689 A CN 105808689A CN 201610119650 A CN201610119650 A CN 201610119650A CN 105808689 A CN105808689 A CN 105808689A
- Authority
- CN
- China
- Prior art keywords
- attribute
- similarity
- transition matrix
- pair
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/29—Geographical information databases
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Remote Sensing (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides a drainage system entity semantic similarity measurement method based on an artificial neural network. From body concept attributes, a space entity semantic similarity measurement model based on the artificial neural network is adopted by aiming at the characteristics of semantics, the model obtains expert sample data through statistics, each group of characteristic similarity is taken as the input parameter of the neural network, an entity semantic similarity given by an expert is taken as an estimated output parameter, a transfer matrix is obtained by training, and the transfer matrix is used for calculating the entity semantic similarity which needs to be measured.
Description
Technical field
The present invention relates to a kind of water system Entity Semantics method for measuring similarity based on artificial neural network, belong to the aspects such as artificial intelligence, geographical space semanteme, Ontological concept, possession reason information science field.
Background technology
At area of geographic information, Semantic Similarity receives more and more attention and studies, be widely used knowledge excavation, information retrieval and comprehensive etc. in, geographic information services semantic matches in Geographical Information Sciences and in geographic information service discovery Semantic Similarity calculating effect more prominent.General Semantic Similarity can be divided into the semantic relation model of the feature based that cognitive psychologist proposes, and computerdom propose based on the Semantic Similarity computation model of relative distance in semantic net.Mainly provide better information retrieval service in GIS area research spatial entities Semantic Similarity and realize integrated, the interoperability of GIS-Geographic Information System.Along with the further investigation that Semantic Similarity is measured, scholars have been presented for calculating based on the Semantic Similarity of character string, based on the Similarity measures of distance, feature based calculating, based on many methods such as Ontological concept model calculating.But method there is also some shortcomings simultaneously, based on character string Semantic Similarity computation model owing to there is polysemy and many words one justice phenomenon, solve during metric question limited;Based in the computation model of semantic distance, in the weight value on limit, result treatment is undesirable;Higher based on the similarity weight subjectivity of each attribute in Ontological concept model computation model.
Summary of the invention
In order to solve the deficiencies in the prior art, the invention provides a kind of water system Entity Semantics method for measuring similarity based on artificial neural network, from Ontological concept attribute, the spatial entities Semantic Similarity measurement model based on artificial neural network is adopted for semantic feature, this model obtains expert's sample data by statistics, using every stack features similarity as nerve network input parameter, the Entity Semantics similarity provided using expert is as estimating output parameter, transition matrix is obtained, the Entity Semantics similarity calculating needs tolerance of use by training.
The present invention solves that its technical problem be the technical scheme is that and provide a kind of water system Entity Semantics method for measuring similarity based on artificial neural network, comprise the following steps:
(1) known training dataset by more than 3 pairs semantic concepts to forming, 2 semantic concepts of semantic concept centering form by one group of attribute, the attribute one_to_one corresponding of 2 semantic concepts, each attribute is respectively provided with the value of attribute, and every pair of semantic concept is respectively provided with desired output;For every pair of semantic concept pair, perform step (2) to step (5):
(2) determining that training data concentrates the type of every pair of attribute, one or more utilizing in following similarity calculation obtain the similarity of every pair of attribute that training datas are concentrated;
(a) general property similarity calculation:
If a pair attribute is orderly attribute, then below equation is adopted to calculate:
sim(a1,b1)=1-| index (a1)-index(b1)|/m
Wherein, sim (a1,b1) represent orderly attribute a1And b1Similarity, index function is for the index value of computation attribute, and described index value is the numbering that the value place of this attribute is interval, and m represents the number that the value of attribute is interval;
If a pair attribute is unordered attribute, then below equation is adopted to calculate:
Wherein, sim (a2,b2) represent unordered attribute a2And b2Similarity;
(b) enumerated attributes similarity calculation:
Employing below equation calculates:
Wherein, sim (a3,b3) represent enumerated attributes a3And b3Similarity, | a3∩b3| represent the number intersecting element in enumerated attributes, | a3∪b3| represent the number of union element in enumerated attributes;
(c) structure attribute similarity calculation:
Employing below equation calculates:
Wherein, sim (a4,b4) represent enumerated attributes a4And b4Similarity, VaAnd VbFor Concept Vectors, Va×VbRepresent Concept Vectors VaAnd VbCross product, Va·VbRepresent Concept Vectors VaAnd VbDot product;
(d) Numeric Attributes similarity calculation:
Employing below equation calculates:
sim(a5,b5)=(va-vb)/λ
Wherein, sim (a5,b5) represent Numeric Attributes a5And b5Similarity, vaAnd vbRepresent a respectively5And b5Property value, λ represents the threshold value of setting;
(3) similarity of step (2) calculated every pair of attribute is formed various dimensions vector (I1,I2,...,In), n is the number of the attribute of semantic concept, the Noumenon property similarity of described various dimensions vector and semantic concept pair;By D={I1,I2,...,InAs input layer I;The interstitial content of hidden layer H is set;
(4) if transition matrix W and transition matrix V initializes, then initialize transition matrix W and transition matrix V, make the line number of transition matrix W and the interstitial content of the interstitial content of columns respectively hidden layer H and output layer O, the interstitial content of the nodes n and hidden layer H of the line number of transition matrix V and columns respectively input layer I;The value of each element of transition matrix W and transition matrix V is initialized as the random number between-1 to 1;Otherwise, step (5) is selected the transition matrix W returned and transition matrix V;
(5) each element in D is performed following steps successively:
(5-1) following steps are performed with propagated forward, to obtain hidden layer H and output layer O:
Wherein, functionHjFor the jth node of hidden layer H, OiI-th node for output layer O;
(5-2) following steps are performed with back-propagating:
W=W+ Δ W
V=V+ Δ V
Wherein, Δ W represents the variable quantity by regulating transition matrix W after back-propagating, Δ WijRepresenting the element in Δ W, Δ V represents the variable quantity by regulating transition matrix V after back-propagating, Δ VijRepresent the element in Δ V, CiRepresenting that training sample concentrates the desired output of i-th pair semantic concept, α is the regulatory factor arranged;
(5-3) each mean square exported between desired output is calculated by below equation poor:
If E than arrange best-minimum-error is little, then neutral net completes training, returns transition matrix W and transition matrix V as two training matrix, is exported by output layer O;Otherwise, return step (5-2) and continue back-propagating to continue training neutral net.
(6) every pair of semantic concept is to after execution step (2) to step (5), and output layer O is final water system Entity Semantics measuring similarity result.
Described regulatory factor α ranges for (0.1,0.8).
Described best-minimum-error range for (0,0.1).
In step (3), the interstitial content of hidden layer is less than the number that N-1, N are that training sample concentrates semantic concept pair.
Before step (5), maximum step number is set, then in step (5-3), returned step (5-2) in the past, judge that what whether recycle time reached maximum step number arranges value, if not up to, then return step (5-2) and continue back-propagating to continue training neutral net, if reached, then neutral net completes training, returns transition matrix W and transition matrix V as two training matrix, is exported by output layer O.
After step (6) obtains final water system Entity Semantics measuring similarity result, the similarity of each attribute of every pair of semantic concept in calculating test data set, itself and final water system Entity Semantics measuring similarity result are compared, to verify the correctness of final water system Entity Semantics measuring similarity result.
The present invention is had advantageous effect in that based on its technical scheme:
(1) present invention is provided with regulatory factor α, regulatory factor α and learning rate, it is possible to transformation matrix control neural network pace of learning when regulating every single-step iteration by arranging learning rate;
(2) present invention is provided with mean square difference E and carrys out the quality that metric learning rate sets, thus avoiding the unstable phenomenon of study;
(3) if hidden layer number is relatively big, and network then there will be over-fitting, the less then similarity of number differs relatively big with actual value, and the hidden layer number of the present invention arranges rationally, it is possible to efficiently complete water system Entity Semantics measuring similarity process;
(4) present invention is after the final water system Entity Semantics measuring similarity result obtaining training dataset, also utilizes the test data set not possessing desired output that result is verified, it is ensured that the correctness of method and the effectiveness of neutral net;
(5) present invention account for the subjective impact of each Noumenon property similarity weights, fully simulates mankind's process in similar cognition, solves Semantics of Space Similarity Problem.
Accompanying drawing explanation
Fig. 1 is neutral net level figure.
Fig. 2 is the spatial form schematic diagram of structural type attribute.
Fig. 3 is the cycle schematic diagram of structural type attribute.
Fig. 4 is learning rate and training iterative steps relation schematic diagram.
Fig. 5 is hidden layer interstitial content and the relation schematic diagram of error (MSE).
Fig. 6 is result of calculation and expertise Comparative result schematic diagram.
Detailed description of the invention
Below in conjunction with drawings and Examples, the invention will be further described.
The invention provides a kind of water system Entity Semantics method for measuring similarity based on artificial neural network, comprise the following steps:
(1) known training dataset by more than 3 pairs semantic concepts to forming, 2 semantic concepts of semantic concept centering form by one group of attribute, the attribute one_to_one corresponding of 2 semantic concepts, each attribute is respectively provided with the value of attribute, every pair of semantic concept is respectively provided with desired output, it is desirable to output is the expectation similarity of expert evaluation;For every pair of semantic concept pair, perform step (2) to step (5):
(2) determining that training data concentrates the type of every pair of attribute, one or more utilizing in following similarity calculation obtain the similarity of every pair of attribute that training datas are concentrated;
With reference to table 1, it it is the Noumenon property information table of a pair semantic concept pair.
Table 1 " small reservoir " and " small lakes " Noumenon property information table
Wherein, " small reservoir " and " small lakes " is semantic concept, and they are respectively provided with 9 attributes: materiality, the origin cause of formation, shape, periodically, size, integrity, function, locus and life cycle, each attribute is respectively provided with the value of attribute.
" size " judges from the value of its attribute, for the orderly attribute in general property, then adopts the model in following (a) to be calculated, and the value of the attribute of the attribute " size " of " small reservoir " is 100,000~10,000,000m3It always has an interval, then interval number is 1, and the numbering in this interval, value place is also 1, in like manner can obtaining the numbering that the interval of the value of the attribute of the attribute " size " of " small lakes ", interval number and place are interval, the similarity that can obtain " size " attribute according to formula is 1.
" materiality " and " origin cause of formation " judges from the value of its attribute, for the unordered attribute in general property, then adopts the model in following (a) to be calculated.
" integrity " and " functional " is enumerated attributes, then adopt the model in following (b) to be calculated, and can obtain similarity respectively 0.2 and 0.75.
" spatial form ", " periodically ", " locus " and " life cycle " is structure attribute, wherein the structure of spatial form is as shown in Figure 2, periodic structure is as it is shown on figure 3, " spatial form ", " periodically ", " locus " and " life cycle " all can be calculated by the model in following (c).According to structure attribute computation model, set regulatory factor here as α=0.5.In order to calculate the similarity of " spatial form ", according to the degree of depth of the level concept figure of Noumenon property and density, the Concept Vectors V of " small reservoir " can be respectively obtained by existing Concept Vectors methodGThe Concept Vectors V of [100040000000000112*0.52*0.5] and " small lakes "p[1000400011000000000], can calculate the similarity obtaining groove-dell according to formula be 0.122, the similarity of dell-dell is 1, then the two average after, the similarity of the structural type attribute " spatial form " of " little reservoir " and " small lakes " is (1+0.122)/2=0.56, and in like manner can calculate the similarity obtaining " periodically " is 0.968.
(a) general property similarity calculation:
If a pair attribute is orderly attribute, then below equation is adopted to calculate:
sim(a1,b1)=1-| index (a1)-index(b1)|/m
Wherein, sim (a1,b1) represent orderly attribute a1And b1Similarity, index function is for the index value of computation attribute, and described index value is the numbering that the value place of this attribute is interval, and m represents the number that the value of attribute is interval;
If a pair attribute is unordered attribute, then below equation is adopted to calculate:
Wherein, sim (a2,b2) represent unordered attribute a2And b2Similarity;
(b) enumerated attributes similarity calculation:
Employing below equation calculates:
Wherein, sim (a3,b3) represent enumerated attributes a3And b3Similarity, | a3∩b3| represent the number intersecting element in enumerated attributes, | a3∪b3| represent the number of union element in enumerated attributes;
(c) structure attribute similarity calculation:
Employing below equation calculates:
Wherein, sim (a4,b4) represent enumerated attributes a4And b4Similarity, VaAnd VbFor Concept Vectors, Va×VbRepresent Concept Vectors VaAnd VbCross product, Va·VbRepresent Concept Vectors VaAnd VbDot product;
(d) Numeric Attributes similarity calculation:
Employing below equation calculates:
sim(a5,b5)=(va-vb)/λ
Wherein, sim (a5,b5) represent Numeric Attributes a5And b5Similarity, vaAnd vbRepresent a respectively5And b5Property value, λ represents the threshold value of setting;
(3) similarity of step (2) calculated every pair of attribute is formed various dimensions vector (I1,I2,...,In), n is the number of the attribute of semantic concept, the Noumenon property similarity of described various dimensions vector and semantic concept pair;By D={I1,I2,...,InAs input layer I;The interstitial content of hidden layer H is set;
(4) if transition matrix W and transition matrix V initializes, then initialize transition matrix W and transition matrix V, make the line number of transition matrix W and the interstitial content of the interstitial content of columns respectively hidden layer H and output layer O, the interstitial content of the nodes n and hidden layer H of the line number of transition matrix V and columns respectively input layer I;The value of each element of transition matrix W and transition matrix V is initialized as the random number between-1 to 1;Otherwise, step (5) is selected the transition matrix W returned and transition matrix V;
The input layer number of neutral net requires consistent with attribute number n.In the present embodiment, neutral net is the final similarity that n the attributes similarity by semantic concept releases semantic concept, final result is an exact numerical values recited, and therefore output layer O node number here is 1 (the namely final similarity between a pair semantic concept);Fig. 1 gives general neutral net schematic diagram, and when the final result of semantic concept is multidimensional, the interstitial content of output layer O can more than 1;
(5) each element in D is performed following steps successively:
(5-1) following steps are performed with propagated forward, to obtain hidden layer H and output layer O:
Wherein, functionHjFor the jth node of hidden layer H, OiFor the i-th node of output layer O, k represents the call number of output layer node;
(5-2) following steps are performed with back-propagating:
W=W+ Δ W
V=V+ Δ V
Wherein, Δ W represents the variable quantity by regulating transition matrix W after back-propagating, Δ WijRepresenting the element in Δ W, Δ V represents the variable quantity by regulating transition matrix V after back-propagating, Δ VijRepresent the element in Δ V, CiRepresenting that training sample concentrates the desired output of i-th pair semantic concept, α is the regulatory factor arranged;
(5-3) each mean square exported between desired output is calculated by below equation poor:
If E than arrange best-minimum-error is little, then neutral net completes training, returns transition matrix W and transition matrix V as two training matrix, is exported by output layer O;Otherwise, return step (5-2) and continue back-propagating to continue training neutral net.
(6) every pair of semantic concept is to after execution step (2) to step (5), and output layer O is final water system Entity Semantics measuring similarity result.
Described regulatory factor α ranges for (0.1,0.8).
Described best-minimum-error range for (0,0.1).
In step (3), the interstitial content of hidden layer is less than the number that N-1, N are that training sample concentrates semantic concept pair.
Before step (5), maximum step number is set, then in step (5-3), returned step (5-2) in the past, judge that what whether recycle time reached maximum step number arranges value, if not up to, then return step (5-2) and continue back-propagating to continue training neutral net, if reached, then neutral net completes training, returns transition matrix W and transition matrix V as two training matrix, is exported by output layer O.
BP artificial neural network is trained by 28 pairs of training samples being concentrated by training sample (semantic concept to), table 2 gives the initial parameter value of BP network, parameter 1 represents the maximum step number of training, parameter 2 represents best-minimum-error, parameter 3 represents every 100 times and shows result, parameter 4 represents the learning rate α of network, and parameter 5 represents the factor of momentum of BP network.
Table 2 training parameter
Training iterations is more many, it is desirable to the difference of output and actual output is more little, closer to the error set, learning rate is excessive, network is it is possible that wild effect, and on the contrary, learning rate may be less likely to cause longer learning time, Fig. 4 gives the graph of a relation of learning rate and train epochs, from graph of a relation it can be seen that when learning rate is less than 0.5, train epochs reduces with this learning rate and reduces, when learning rate is more than 0.6, training iterative steps is relatively stable.It can be seen that when learning rate is less than 0.5, network is very sensitive to learning rate, when learning rate is more than 0.6, training iterative steps is affected less by learning rate, based on above analysis, it is possible to learning rate is set as 0.5.
It is also extremely important for setting up the setting of hidden layer interstitial content in process at BP network, not only directly affects the performance of network, and it would furthermore be possible to directly result in network training " over-fitting " phenomenon occur.Fig. 5 gives training error and the graph of a relation of concealed nodes number, and when nodes is less than 19, error increases along with nodes and reduces.When interstitial content is relatively stable more than 21 time errors.In order to avoid Expired Drugs occurring and guaranteeing the expressive ability of network, it is possible to the hidden layer interstitial content of setting network is 21.
After step (6) obtains final water system Entity Semantics measuring similarity result, the similarity of each attribute of every pair of semantic concept in calculating test data set, itself and final water system Entity Semantics measuring similarity result are compared, to verify the correctness of final water system Entity Semantics measuring similarity result:
Test concept data are five groups of data such as hondo-seasonal river, spring-well, reservoir-lake, the first Noumenon property similarity between our computational entity semanteme, and result of calculation is as the network importation being activated, and output is then the similarity of Entity Semantics.Fig. 6 gives expertise result, the inventive method and the Entity Semantics similarity based on Concept Vectors algorithm, it can be seen that the BP artificial neural network result trained by the inventive method is closer to the thinking process of the mankind.
Claims (6)
1. the water system Entity Semantics method for measuring similarity based on artificial neural network, it is characterised in that comprise the following steps:
(1) known training dataset by more than 3 pairs semantic concepts to forming, 2 semantic concepts of semantic concept centering form by one group of attribute, the attribute one_to_one corresponding of 2 semantic concepts, each attribute is respectively provided with the value of attribute, and every pair of semantic concept is respectively provided with desired output;For every pair of semantic concept pair, perform step (2) to step (5):
(2) determining that training data concentrates the type of every pair of attribute, one or more utilizing in following similarity calculation obtain the similarity of every pair of attribute that training datas are concentrated;
(a) general property similarity calculation:
If a pair attribute is orderly attribute, then below equation is adopted to calculate:
sim(a1,b1)=1-| index (a1)-index(b1)|/m
Wherein, sim (a1,b1) represent orderly attribute a1And b1Similarity, index function is for the index value of computation attribute, and described index value is the numbering that the value place of this attribute is interval, and m represents the number that the value of attribute is interval;
If a pair attribute is unordered attribute, then below equation is adopted to calculate:
Wherein, sim (a2,b2) represent unordered attribute a2And b2Similarity;
(b) enumerated attributes similarity calculation:
Employing below equation calculates:
Wherein, sim (a3,b3) represent enumerated attributes a3And b3Similarity, | a3∩b3| represent the number intersecting element in enumerated attributes, | a3∪b3| represent the number of union element in enumerated attributes;
(c) structure attribute similarity calculation:
Employing below equation calculates:
Wherein, sim (a4,b4) represent enumerated attributes a4And b4Similarity, VaAnd VbFor Concept Vectors, Va×VbRepresent Concept Vectors VaAnd VbCross product, Va·VbRepresent Concept Vectors VaAnd VbDot product;
(d) Numeric Attributes similarity calculation:
Employing below equation calculates:
sim(a5,b5)=(va-vb)/λ
Wherein, sim (a5,b5) represent Numeric Attributes a5And b5Similarity, vaAnd vbRepresent a respectively5And b5Property value, λ represents the threshold value of setting;
(3) similarity of step (2) calculated every pair of attribute is formed various dimensions vector (I1,I2,...,In), n is the number of the attribute of semantic concept, the Noumenon property similarity of described various dimensions vector and semantic concept pair;By D={I1,I2,...,InAs input layer I;The interstitial content of hidden layer H is set;
(4) if transition matrix W and transition matrix V initializes, then initialize transition matrix W and transition matrix V, make the line number of transition matrix W and the interstitial content of the interstitial content of columns respectively hidden layer H and output layer O, the interstitial content of the nodes n and hidden layer H of the line number of transition matrix V and columns respectively input layer I;The value of each element of transition matrix W and transition matrix V is initialized as the random number between-1 to 1;Otherwise, step (5) is selected the transition matrix W returned and transition matrix V;
(5) each element in D is performed following steps successively:
(5-1) following steps are performed with propagated forward, to obtain hidden layer H and output layer O:
Wherein, functionHjFor the jth node of hidden layer H, OiI-th node for output layer O;
(5-2) following steps are performed with back-propagating:
W=W+ Δ W
V=V+ Δ V
Wherein, Δ W represents the variable quantity by regulating transition matrix W after back-propagating, Δ WijRepresenting the element in Δ W, Δ V represents the variable quantity by regulating transition matrix V after back-propagating, Δ VijRepresent the element in Δ V, CiRepresenting that training sample concentrates the desired output of i-th pair semantic concept, α is the regulatory factor arranged;
(5-3) each mean square exported between desired output is calculated by below equation poor:
If E than arrange best-minimum-error is little, then neutral net completes training, returns transition matrix W and transition matrix V as two training matrix, is exported by output layer O;Otherwise, return step (5-2) and continue back-propagating to continue training neutral net.
(6) every pair of semantic concept is to after execution step (2) to step (5), and output layer O is final water system Entity Semantics measuring similarity result.
2. the water system Entity Semantics method for measuring similarity based on artificial neural network according to claim 1, it is characterised in that: described regulatory factor α ranges for (0.1,0.8).
3. the water system Entity Semantics method for measuring similarity based on artificial neural network according to claim 1, it is characterised in that: described best-minimum-error range for (0,0.1).
4. the water system Entity Semantics method for measuring similarity based on artificial neural network according to claim 1, it is characterised in that: in step (3), the interstitial content of hidden layer is less than the number that N-1, N are that training sample concentrates semantic concept pair.
5. the water system Entity Semantics method for measuring similarity based on artificial neural network according to claim 1, it is characterized in that: before step (5), maximum step number is set, then in step (5-3), returned step (5-2) in the past, judge that what whether recycle time reached maximum step number arranges value, if not up to, then return step (5-2) and continue back-propagating to continue training neutral net, if reached, then neutral net completes training, return transition matrix W and transition matrix V as two training matrix, output layer O is exported.
6. the water system Entity Semantics method for measuring similarity based on artificial neural network according to claim 1, it is characterized in that: after step (6) obtains final water system Entity Semantics measuring similarity result, the similarity of each attribute of every pair of semantic concept in calculating test data set, itself and final water system Entity Semantics measuring similarity result are compared, to verify the correctness of final water system Entity Semantics measuring similarity result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610119650.7A CN105808689A (en) | 2016-03-03 | 2016-03-03 | Drainage system entity semantic similarity measurement method based on artificial neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610119650.7A CN105808689A (en) | 2016-03-03 | 2016-03-03 | Drainage system entity semantic similarity measurement method based on artificial neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN105808689A true CN105808689A (en) | 2016-07-27 |
Family
ID=56466442
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610119650.7A Pending CN105808689A (en) | 2016-03-03 | 2016-03-03 | Drainage system entity semantic similarity measurement method based on artificial neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105808689A (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107783958A (en) * | 2016-08-31 | 2018-03-09 | 科大讯飞股份有限公司 | A kind of object statement recognition methods and device |
CN108509651A (en) * | 2018-04-17 | 2018-09-07 | *** | The distributed approximation searching method with secret protection based on semantic consistency |
CN108595506A (en) * | 2018-03-21 | 2018-09-28 | 上海数据交易中心有限公司 | Demand matching process and device, storage medium, terminal |
CN109189941A (en) * | 2018-09-07 | 2019-01-11 | 百度在线网络技术(北京)有限公司 | For updating the method, apparatus, equipment and medium of model parameter |
CN109344405A (en) * | 2018-09-25 | 2019-02-15 | 艾凯克斯(嘉兴)信息科技有限公司 | A kind of similarity processing method based on TF-IDF thought and neural network |
CN110134965A (en) * | 2019-05-21 | 2019-08-16 | 北京百度网讯科技有限公司 | Method, apparatus, equipment and computer readable storage medium for information processing |
CN110232722A (en) * | 2019-06-13 | 2019-09-13 | 腾讯科技(深圳)有限公司 | A kind of image processing method and device |
CN111782762A (en) * | 2020-05-12 | 2020-10-16 | 北京三快在线科技有限公司 | Method and device for determining similar questions in question answering application and electronic equipment |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102156726A (en) * | 2011-04-01 | 2011-08-17 | 中国测绘科学研究院 | Geographic element querying and extending method based on semantic similarity |
CN103092828A (en) * | 2013-02-06 | 2013-05-08 | 杭州电子科技大学 | Text similarity measuring method based on semantic analysis and semantic relation network |
CN104699763A (en) * | 2015-02-11 | 2015-06-10 | 中国科学院新疆理化技术研究所 | Text similarity measuring system based on multi-feature fusion |
-
2016
- 2016-03-03 CN CN201610119650.7A patent/CN105808689A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102156726A (en) * | 2011-04-01 | 2011-08-17 | 中国测绘科学研究院 | Geographic element querying and extending method based on semantic similarity |
CN103092828A (en) * | 2013-02-06 | 2013-05-08 | 杭州电子科技大学 | Text similarity measuring method based on semantic analysis and semantic relation network |
CN104699763A (en) * | 2015-02-11 | 2015-06-10 | 中国科学院新疆理化技术研究所 | Text similarity measuring system based on multi-feature fusion |
Non-Patent Citations (1)
Title |
---|
YONGYANG XU ET AL.: "Research on Semantics of Entity Space Similarity Measure Based on Artificial Neural Networks", 《2015 23RD INTERNATIONAL CONFERENCE ON GEOINFORMATICS》 * |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107783958A (en) * | 2016-08-31 | 2018-03-09 | 科大讯飞股份有限公司 | A kind of object statement recognition methods and device |
CN107783958B (en) * | 2016-08-31 | 2021-07-02 | 科大讯飞股份有限公司 | Target statement identification method and device |
CN108595506B (en) * | 2018-03-21 | 2020-11-27 | 上海数据交易中心有限公司 | Demand matching method and device, storage medium and terminal |
CN108595506A (en) * | 2018-03-21 | 2018-09-28 | 上海数据交易中心有限公司 | Demand matching process and device, storage medium, terminal |
CN108509651A (en) * | 2018-04-17 | 2018-09-07 | *** | The distributed approximation searching method with secret protection based on semantic consistency |
CN108509651B (en) * | 2018-04-17 | 2019-03-12 | *** | The distributed approximation searching method with secret protection based on semantic consistency |
CN109189941A (en) * | 2018-09-07 | 2019-01-11 | 百度在线网络技术(北京)有限公司 | For updating the method, apparatus, equipment and medium of model parameter |
CN109344405A (en) * | 2018-09-25 | 2019-02-15 | 艾凯克斯(嘉兴)信息科技有限公司 | A kind of similarity processing method based on TF-IDF thought and neural network |
CN109344405B (en) * | 2018-09-25 | 2023-04-14 | 艾凯克斯(嘉兴)信息科技有限公司 | Similarity processing method based on TF-IDF thought and neural network |
CN110134965A (en) * | 2019-05-21 | 2019-08-16 | 北京百度网讯科技有限公司 | Method, apparatus, equipment and computer readable storage medium for information processing |
CN110134965B (en) * | 2019-05-21 | 2023-08-18 | 北京百度网讯科技有限公司 | Method, apparatus, device and computer readable storage medium for information processing |
CN110232722A (en) * | 2019-06-13 | 2019-09-13 | 腾讯科技(深圳)有限公司 | A kind of image processing method and device |
CN110232722B (en) * | 2019-06-13 | 2023-08-04 | 腾讯科技(深圳)有限公司 | Image processing method and device |
CN111782762A (en) * | 2020-05-12 | 2020-10-16 | 北京三快在线科技有限公司 | Method and device for determining similar questions in question answering application and electronic equipment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105808689A (en) | Drainage system entity semantic similarity measurement method based on artificial neural network | |
CN109002917A (en) | Total output of grain multidimensional time-series prediction technique based on LSTM neural network | |
CN111291556B (en) | Chinese entity relation extraction method based on character and word feature fusion of entity meaning item | |
CN105279554B (en) | The training method and device of deep neural network based on Hash coding layer | |
CN113190688B (en) | Complex network link prediction method and system based on logical reasoning and graph convolution | |
CN102495919B (en) | Extraction method for influence factors of carbon exchange of ecosystem and system | |
CN104598611A (en) | Method and system for sequencing search entries | |
CN104866578A (en) | Hybrid filling method for incomplete data | |
CN106528610A (en) | Knowledge graph representation learning method based on path tensor decomposition | |
CN108647839A (en) | Voltage-stablizer water level prediction method based on cost-sensitive LSTM Recognition with Recurrent Neural Network | |
CN103593703A (en) | Neural network optimizing system and method based on genetic algorithms | |
CN105787557A (en) | Design method of deep nerve network structure for computer intelligent identification | |
CN107705556A (en) | A kind of traffic flow forecasting method combined based on SVMs and BP neural network | |
CN109242223A (en) | The quantum support vector machines of city Public Buildings Fire Risk is assessed and prediction technique | |
CN112232526B (en) | Geological disaster vulnerability evaluation method and system based on integration strategy | |
CN105760673A (en) | Fluvial facies reservoir earthquake sensitive parameter template analysis method | |
CN106779053A (en) | The knowledge point of a kind of allowed for influencing factors and neutral net is known the real situation method | |
CN107423820A (en) | The knowledge mapping of binding entity stratigraphic classification represents learning method | |
CN103426027A (en) | Intelligent normal pool level optimal selection method based on genetic neural network models | |
CN108182500A (en) | Ammunition Storage Reliability Forecasting Methodology based on accelerated life test | |
CN105809201A (en) | Identification method and device for autonomously extracting image meaning concepts in biologically-inspired mode | |
CN112578089B (en) | Air pollutant concentration prediction method based on improved TCN | |
CN106127330A (en) | Fluctuating wind speed Forecasting Methodology based on least square method supporting vector machine | |
CN113590799A (en) | Weak supervision knowledge graph question-answering method based on multi-view reasoning | |
CN103440275A (en) | Prim-based K-means clustering method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160727 |
|
RJ01 | Rejection of invention patent application after publication |