CN110727833B - Multi-view learning-based graph data retrieval result optimization method - Google Patents

Multi-view learning-based graph data retrieval result optimization method Download PDF

Info

Publication number
CN110727833B
CN110727833B CN201910944120.XA CN201910944120A CN110727833B CN 110727833 B CN110727833 B CN 110727833B CN 201910944120 A CN201910944120 A CN 201910944120A CN 110727833 B CN110727833 B CN 110727833B
Authority
CN
China
Prior art keywords
graph data
data
retrieval
graph
view
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
Application number
CN201910944120.XA
Other languages
Chinese (zh)
Other versions
CN110727833A (en
Inventor
钟昊文
刘波
肖燕珊
林志全
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangdong University of Technology
Original Assignee
Guangdong University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Guangdong University of Technology filed Critical Guangdong University of Technology
Priority to CN201910944120.XA priority Critical patent/CN110727833B/en
Publication of CN110727833A publication Critical patent/CN110727833A/en
Application granted granted Critical
Publication of CN110727833B publication Critical patent/CN110727833B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/901—Indexing; Data structures therefor; Storage structures
    • G06F16/9024—Graphs; Linked lists
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/903—Querying
    • G06F16/90335—Query processing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/903—Querying
    • G06F16/9038—Presentation of query results
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/906—Clustering; Classification
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention relates to a method for optimizing a graph data retrieval result based on multi-view learning, which comprises the following steps: s1: generating training data according to click data during user retrieval, and constructing a plurality of visual angles of the training data; s2: setting parameters of each visual angle of the training data to improve the classification performance of the training data; s3: constructing a graph data retrieval evaluation model; and optimizing the same; s4: calculating the grade of the graph data in the database by using the evaluation model; s5: and sequencing the graph data according to the scores to obtain a retrieval result. According to the invention, the click data can be utilized to improve the condition of poor accuracy of the current image data retrieval, and the retrieval quality is improved; the target model provided by the invention considers the retrieval problem of processing graph data from multiple angles. By simultaneously utilizing a plurality of feature mappings of the graph data and adding multi-view constraints, the consensus and complementation principle in multi-view learning is met, and the accuracy of the model for graph data retrieval is improved.

Description

Graph data retrieval result optimization method based on multi-view learning
Technical Field
The invention relates to the field of machine learning, in particular to a method for optimizing a graph data retrieval result based on multi-view learning.
Background
Graph data is an abstract data structure, consisting of vertices and edges. Data in many fields can be described or modeled using graph data, such as DNA in biology, chemical compounds and molecular structures in chemistry, network structures in computers, and so forth. With the rapid development of the big data era, many new types of data are also described using graph data, such as social networks, knowledge graphs, and the like. The graph data is receiving more and more attention because of its strong expressive power. Graph data may be used to describe a particular relationship between something, with vertices representing things and lines connecting two points representing a relationship between two corresponding things.
The retrieval of graph data refers to finding out similar graph data from a graph database for a query of one graph data. The conventional data retrieval method is very suitable for retrieving data with a general structure, but the general data retrieval method cannot be used for retrieving graph data due to the special structure of the graph data. In the existing graph data retrieval research, the most common graph data retrieval methods have three types: connectivity retrieval, (II) graph matching retrieval, and (III) keyword retrieval. The connectivity retrieval is mainly carried out by checking whether a node u in the graph data has a path communicated with the node v or not, so as to retrieve the graph data; the graph matching retrieval is mainly implemented by calculating the matching degree of two graph data to retrieve the graph data by using whether points in the two graph data have the same label and whether edges communicated with mutually corresponding points in two point sets correspond as a judgment standard; keyword retrieval finds graph data from a database that is relevant to a given keyword, primarily by the given keyword or keywords. These methods still have many problems, such as poor quality of retrieval, the retrieved result is not our target result, and so on.
In order to realize the retrieval function, retrieval methods such as connectivity retrieval, graph matching retrieval, keyword retrieval and the like can be used, but the retrieval methods only consider a certain experience angle, do not consider the angle of a user, and do not consider the problems of various potential characteristics and complex structure of graph data. Therefore, the methods have low retrieval accuracy for users.
Disclosure of Invention
The invention provides a method for optimizing an image data retrieval result based on multi-view learning, aiming at overcoming the defect of poor quality of image data retrieval in the prior art.
The common graph data retrieval optimization method does not consider utilizing the existing retrieval results, and when a user retrieves graph data, the user generally only has interest in a part of the graph data in the retrieval results, so that click data during user retrieval can be used as training data to optimize the retrieval quality of the graph data. Meanwhile, the scale of the graph data is increasingly huge, and the relationship represented by the graph data is also increasingly complex, but most of the existing graph data retrieval methods only study the situation that the graph data is described from one angle, which is called single view learning (single view learning), and the feature of the graph data of a single view is one-sided. In practice, we can describe the object from multiple angles, each of which describes the data in a separate way. One way to improve performance by exploiting the diversity of different views of the graph data is called multi-view learning (multi-view learning).
For feature extraction of graph data, the method is divided into two categories: the method comprises (I) a feature mining based method (feature mining based methods) and (II) an embedding learning based method (embedding learning based methods). Firstly, finding topological features or sub-graph features based on a feature mining method, and then, representing each graph data by a vector consisting of 0 and 1 according to whether the graph data contains corresponding features; the method based on embedded learning is expected to maximally retain structural features and attributes of graph data in the process of converting the graph data into low-dimensional vectors. Therefore, different graph data feature extraction methods may mine or learn features of graph data in a manner unique to themselves.
Therefore, a better retrieval result can be obtained by constructing a plurality of visual angles of the graph data through different graph data feature extraction methods to carry out optimization of graph data retrieval. The method utilizes click data during user retrieval, constructs a plurality of visual angles of the graph data through a plurality of different graph data feature extraction methods, and meets the consensus and complementation principles in multi-view learning, thereby optimizing the retrieval result of the graph data.
The method comprises the following steps:
s1: generating training data according to click data retrieved by a user, and constructing a plurality of visual angles of the training data;
s2: setting parameters of each visual angle of the training data to improve the classification performance of the training data;
s3: constructing a graph data retrieval evaluation model; and optimizing the same;
s4: calculating the score of the graph data in the database by using the evaluation model;
s5: and sequencing the graph data according to the scores to obtain a retrieval result.
Preferably, S1 comprises the steps of:
s1.1: extracting corresponding graph data according to click data during user retrieval, marking the graph data as regular graph data, and marking other graph data as non-label graph data;
then n can be obtained p Collection of personal schematic data
Figure BDA0002223708660000031
And n u Collection of individual label-free graph data
Figure BDA0002223708660000032
Each graph data G has a class label Y, Y belongs to Y = { +1,0, -1}, and the labels of the graph data are positive, non-labeled and negative;
s1.2: graph data sample G by utilizing graph data feature extraction method i Mapping to v feature vectors
Figure BDA0002223708660000033
(v) Representing the v extraction method, for the graph data sample G i There are m number of viewing angles (based on m number of types of extraction method;)>
Figure BDA0002223708660000034
Preferably, the parameter setting in S2 is specifically: setting a weight γ for each view angle v ,γ v Significance of v views of the reaction map data; setting constraint weights
Figure BDA0002223708660000035
C 2 ，/>
Figure BDA0002223708660000036
Weights representing ordering constraints, C 2 Weights representing a multi-view constraint; setting a regularization parameter epsilon, wherein the parameter epsilon is used for allowing a small part of the graph data to violate the constraint;
preferably, S3 is specifically: v feature vectors
Figure BDA0002223708660000037
As an input to a multi-view ordering Support Vector Machine (SVM), wherein the SVM is made up of a plurality of different weight vectors w v And (4) forming.
Preferably, the target equation of the graph data retrieval and evaluation model in S3 is:
Figure BDA0002223708660000038
Figure BDA0002223708660000039
/>
Figure BDA00022237086600000310
Figure BDA00022237086600000311
R＝{(p,u):G p ∈P,G u ∈U},
a＝1,…,m-1,b＝a+1,…,m,
k＝1,…,n,v＝1,…,m
wherein, the first and the second end of the pipe are connected with each other,
Figure BDA00022237086600000312
representing the characteristics of the graph data p in the v-th view angle; />
Figure BDA00022237086600000313
Representing the characteristics of the graph data u in the v-th view; />
Figure BDA00022237086600000314
Representing the characteristics of the graph data k in the v-th view angle; p, u, k are subscripts, P is used to indicate that the graph data P is in the set P; u is used to represent graph data U in the set U; k =1 … n for representationEach graph data in the set of graph data;
w v representing the weight vector in the v-th view,
Figure BDA00022237086600000315
to relax the variable, which is used to allow a small portion of the graph data to violate the constraints, we need to minimize the variable.
ε represents the regularization parameter, set by the user.
Each of the normal map data p respectively forms a map data pair (p, u) with each of the unlabeled map data u, and R represents a set of these map data pairs.
First constraint
Figure BDA0002223708660000041
The score of positive example graph data is always larger than that of negative example graph data when the graph data is scored by the model.
Second constraint
Figure BDA0002223708660000042
The method is used for enabling scores of different view angles of the graph data to be consistent as much as possible when the graph data is scored by our model.
Preferably, the optimization of the graph data retrieval evaluation model in S3 is:
the dual problem is obtained by deducing a target equation of the graph data retrieval evaluation model by using a Lagrange multiplier method, wherein the target equation is related to a non-negative Lagrange multiplier
Figure BDA0002223708660000043
The formula is as follows:
Figure BDA0002223708660000044
Figure BDA0002223708660000045
Figure BDA0002223708660000046
wherein the content of the first and second substances,
Figure BDA0002223708660000047
solving the derived dual problem by using an optimization algorithm (such as an SMO algorithm), and solving the optimal Lagrange multiplier
Figure BDA0002223708660000048
I.e. can be used to calculate w v 。
Preferably, the calculation process of the score of the graph data in S4 is:
for graph data G, its m perspective representations are first constructed
Figure BDA0002223708660000049
The score for graph data G is then calculated using the following formula:
Figure BDA00022237086600000410
compared with the prior art, the technical scheme of the invention has the beneficial effects that:
the target model provided by the invention considers the click data in the existing user retrieval, converts the retrieval optimization problem of the graph data into the sorting problem of the positive sample and the unmarked sample of the graph data, and can improve the condition of poor retrieval accuracy of the current graph data by using the click data and improve the retrieval quality by combining a sorting support vector machine model;
the target model provided by the invention considers the retrieval problem of processing graph data from multiple angles. By simultaneously utilizing a plurality of feature mappings of the graph data and adding multi-view constraints, the consensus and complementation principle in multi-view learning is met, and the accuracy of the model for graph data retrieval is improved.
The invention can optimize the target model while establishing the target model, thereby reducing the complexity.
Drawings
Fig. 1 is a flowchart of a method for optimizing a retrieval result of graph data based on multi-view learning according to embodiment 1.
Fig. 2 is an optimization example diagram.
Detailed Description
The drawings are for illustrative purposes only and are not to be construed as limiting the patent;
for the purpose of better illustrating the present embodiments, certain elements of the drawings may be omitted, enlarged or reduced, and do not represent the size of an actual product;
it will be understood by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
The technical solution of the present invention is further described below with reference to the accompanying drawings and examples.
Example 1
The embodiment provides a method for optimizing a graph data retrieval result based on multi-view learning, as shown in fig. 1, the method includes the following steps:
s1: generating training data according to click data retrieved by a user, and constructing a plurality of visual angles of the training data;
s2: setting parameters of each visual angle of the training data to improve the classification performance of the training data;
s3: constructing a graph data retrieval evaluation model; and optimizing the same;
s4: calculating the score of the graph data in the database by using the evaluation model;
s5: and sequencing the graph data according to the scores to obtain a retrieval result.
In this embodiment, for a given user to retrieve click data, we extract the corresponding graph data and mark it as the positive example graph data, and the other graph data are marked as the unlabeled graph data. Then we can obtain n p Collection of personal schematic data
Figure BDA0002223708660000051
And n u Set of unlabeled graph data->
Figure BDA0002223708660000052
Figure BDA0002223708660000053
Each graph data G has a class label Y, Y ∈ Y = { +1,0, -1}, indicating that the label of the graph data is positive, unlabeled, and negative.
(1) Firstly, the graph data sample G is extracted by different graph data characteristic extraction methods i Mapping to v eigenvectors
Figure BDA0002223708660000054
(v) Represents the extraction method of the v th kind, v feature vectors ≥ are present>
Figure BDA0002223708660000055
As an input to a multi-view order SVM, wherein the multi-view order SVM is formed from a plurality of different vectors W (v) Composition as our assessment model;
(2) After the first stage, the evaluation model may be used to generate a score for the graph data.
(3) And finally, obtaining a retrieval result according to the ranking of the scores.
To better describe the implementation of our proposed new method, we define several operations as follows:
1. setting parameters:
setting a weight γ for each view angle v ,γ v Significance of v views of the reaction map data; setting constraint weights
Figure BDA0002223708660000061
C 2 ，
Figure BDA0002223708660000062
Weight representing ordering constraint, C 2 Weights representing a multi-view constraint; a regularization parameter epsilon is set that allows a small portion of the graph data to violate the constraint. In thatIn this embodiment, the configuration of the hyper-parameters has an effect on the classification effect.
In the implementation process, the empirical selection range of the hyper-parameters can be obtained through cross validation of each data set, but different hyper-parameters are selected for different data sets, which is time-consuming and labor-consuming. Therefore, for convenience, the method selects to optimize the hyper-parameters on one data set, and other data sets can also use the parameter setting, namely, the classification performance can be improved through parameter optimization.
2. Multi-view construction
For multi-view construction, different feature extraction methods can be adopted to map graph data, such as transformation of the graph data into vectors (graph 2 vec), mining of frequent subgraphs (gSpan), and the like.
By using
Figure BDA0002223708660000063
The feature extracted from the graph data i by the v-th feature extraction method is shown. Then for the graph data sample G i There are m number of viewing angles (based on m number of types of extraction method;)>
Figure BDA0002223708660000064
3. Determining an evaluation model
We want the more relevant graph data to the user target ranked the earlier in our ranking results. The sorted support vector machine model is well suited for this purpose, but the general sorted support vector machine model is single-view, so it is expanded.
Since multi-view data is processed, this means that the evaluation function should also follow the consistency and complementarity of multi-view learning. The objective equation of the evaluation model is therefore as follows:
Figure BDA0002223708660000065
Figure BDA0002223708660000066
Figure BDA0002223708660000067
Figure BDA0002223708660000068
R＝{(p,u):G p ∈P,G u ∈U},
a＝1,…,m-1,b＝a+1,…,m,
k＝1,…,n,v＝1,…,m
wherein the content of the first and second substances,
Figure BDA0002223708660000069
representing the characteristics of the graph data p in the v-th view angle; />
Figure BDA00022237086600000610
Representing the characteristics of the graph data u in the v-th view; />
Figure BDA0002223708660000071
Representing the characteristics of the graph data k in the v-th view angle; p, u, k are subscripts, P is used to indicate that the graph data P is in the set P; u is used to represent graph data U in the set U; k =1 … n for representing each graph data in the graph data set;
w v representing the weight vector in the v-th view,
Figure BDA0002223708660000072
to relax a variable, which is used to allow a small portion of the graph data to violate a constraint, we need to minimize the variable.
ε represents the regularization parameter, set by the user.
Each of the normal map data p respectively forms a map data pair (p, u) with each of the unlabeled map data u, and R represents a set of these map data pairs.
First of allA constraint
Figure BDA0002223708660000073
The score of positive example graph data is always larger than that of negative example graph data when the graph data is scored by the model.
Second constraint
Figure BDA0002223708660000074
The method is used for enabling scores of different visual angles of the graph data to be consistent as much as possible when the graph data is scored by the model.
4. Optimization of the objective equation:
deriving the target equation of the model to obtain the dual problem of the model by using a Lagrange multiplier technology, wherein the dual problem is related to a non-negative Lagrange multiplier
Figure BDA0002223708660000075
The formula is as follows:
Figure BDA0002223708660000076
Figure BDA0002223708660000077
Figure BDA0002223708660000078
wherein
Figure BDA0002223708660000079
The derived dual problem can be solved by using an optimization algorithm such as an SMO algorithm, and the solved optimal Lagrange multiplier
Figure BDA00022237086600000710
I.e. can be used to calculate w v 。
5. Calculating a score
For a certain graph data G, we first construct its m view representations
Figure BDA00022237086600000711
Its score can be calculated using the following formula:
Figure BDA00022237086600000712
therefore, the graph data in the database can use the formula to calculate the scores, and the retrieval results can be obtained after the scores are sorted.
As a specific embodiment, as shown in fig. 2, the molecular structure is described by using graph data and a corresponding database is established, the search result sequence of the keyword "anticancer" is graph data 1, graph data 2, graph data 3, graph data 4, graph data 5, and graph data 6, the graph data 2 and graph data 6 clicked by the user 1 in the search result, the graph data 2 and graph data 4 clicked by the user 2 in the search result, the click data of the user 1 and the user 2 is used as a training sample, and after the optimization by the method, the sequence of the search result sequence is changed into graph data 2, graph data 1, graph data 4, graph data 6, graph data 3, and graph data 5.
The terms describing positional relationships in the drawings are for illustrative purposes only and are not to be construed as limiting the patent;
it should be understood that the above-described embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the claims of the present invention.

Claims (3)

1. A method for optimizing a graph data retrieval result based on multi-view learning is characterized by comprising the following steps:
s1: generating training data according to click data retrieved by a user, and constructing a plurality of visual angles of the training data;
s2: setting parameters of each visual angle of the training data to improve the classification performance of the training data;
s3: constructing a graph data retrieval evaluation model; and optimizing the same;
s4: calculating the score of the graph data in the database by using the evaluation model;
s5: sorting the graph data according to the scores to obtain a retrieval result;
s1 comprises the following steps:
s1.1: extracting corresponding graph data according to click data during user retrieval, marking the graph data as regular graph data, and marking other graph data as non-label graph data;
then n can be obtained p Collection of personal schematic data
Figure FDA0004089028640000011
And n u A collection of unlabeled graph data
Figure FDA0004089028640000012
Each graph data G has a class label Y, Y ∈ Y = { +1,0, -1}, indicating that the label of the graph data is positive, unlabeled, and negative;
s1.2: graph data sample G by utilizing graph data feature extraction method i Mapping to v eigenvectors
Figure FDA0004089028640000013
(v) Representing the v extraction method, for the graph data sample G i There are m number of viewing angles (based on m number of types of extraction method;)>
Figure FDA0004089028640000014
The parameter setting in S2 is specifically: setting a weight γ for each view angle v ,γ v Significance of v views of the reaction map data; setting constraint weights
Figure FDA0004089028640000015
Figure FDA0004089028640000016
Weights representing ordering constraints, C 2 Weights representing a multi-view constraint; setting a regularization parameter epsilon, wherein the parameter epsilon is used for allowing a small part of the graph data to violate the constraint;
s3 specifically comprises the following steps: v feature vectors
Figure FDA0004089028640000017
As the input of the multi-view ordering support vector machine, wherein the multi-view ordering SVM is composed of a plurality of different weight vectors w v Composition is carried out;
the target equation of the graph data retrieval evaluation model in S3 is as follows:
Figure FDA0004089028640000018
Figure FDA0004089028640000019
Figure FDA00040890286400000110
Figure FDA0004089028640000021
R＝{(p，u):G p ∈P，G u ∈U}，
a＝1，…，m-1，b＝a+1，…，m，
k＝1，…，n，v＝1，…，m
wherein the content of the first and second substances,
Figure FDA0004089028640000022
representing the characteristics of the graph data p in the v-th view angle; />
Figure FDA0004089028640000023
Representing the characteristics of the graph data u in the v-th view; />
Figure FDA0004089028640000024
Representing the characteristics of the graph data k in the v-th view angle; p, u, k are subscripts, P is used to indicate that the graph data P is in the set P; u is used to represent graph data U in the set U; k =1 … n for representing each graph data in the graph data set; />
w v Representing the weight vector in the v-th view,
Figure FDA0004089028640000025
to relax a variable, which is used to allow a small portion of the graph data to violate a constraint, we need to minimize the variable;
ε represents the regularization parameter;
each positive case diagram data p respectively forms a diagram data pair (p, u) with each unlabeled diagram data u, and R represents a set of the diagram data pairs;
constraining
Figure FDA0004089028640000026
When the model scores the chart data, the score of the positive example chart data is always larger than that of the negative example chart data;
constraining
Figure FDA0004089028640000027
When the model is used for scoring the graph data, scores of different visual angles of the graph data are consistent as much as possible.
2. The method for optimizing the search result of the graph data based on the multi-view learning according to claim 1, wherein the optimization of the graph data search evaluation model in S3 is:
the dual problem is obtained by deducing a target equation of the graph data retrieval evaluation model by using a Lagrange multiplier method, wherein the target equation is related to a non-negative Lagrange multiplier
Figure FDA0004089028640000028
The formula is as follows:
Figure FDA0004089028640000029
Figure FDA00040890286400000210
Figure FDA00040890286400000211
wherein the content of the first and second substances,
Figure FDA00040890286400000212
Figure FDA00040890286400000213
solving the derived dual problem to obtain the optimal Lagrange multiplier
Figure FDA00040890286400000214
I.e. can be used to calculate w v 。
3. The method for optimizing the search result of the graph data based on the multi-view learning of claim 2, wherein the score of the graph data in S4 is calculated by:
for graph data H, it is first constructedm view representations
Figure FDA0004089028640000031
The score for graph data H is then calculated using the following formula:
Figure FDA0004089028640000032
/>
CN201910944120.XA 2019-09-30 2019-09-30 Multi-view learning-based graph data retrieval result optimization method Active CN110727833B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910944120.XA CN110727833B (en) 2019-09-30 2019-09-30 Multi-view learning-based graph data retrieval result optimization method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910944120.XA CN110727833B (en) 2019-09-30 2019-09-30 Multi-view learning-based graph data retrieval result optimization method

Publications (2)

Publication Number Publication Date
CN110727833A CN110727833A (en) 2020-01-24
CN110727833B true CN110727833B (en) 2023-03-31

Family

ID=69218697

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910944120.XA Active CN110727833B (en) 2019-09-30 2019-09-30 Multi-view learning-based graph data retrieval result optimization method

Country Status (1)

Country Link
CN (1) CN110727833B (en)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108021930A (en) * 2017-11-16 2018-05-11 苏州大学 A kind of adaptive multi-view image sorting technique and system
CN108846046A (en) * 2018-05-30 2018-11-20 大连理工大学 The image search method of insertion is kept based on multi-angle of view Partial Reconstruction

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108021930A (en) * 2017-11-16 2018-05-11 苏州大学 A kind of adaptive multi-view image sorting technique and system
CN108846046A (en) * 2018-05-30 2018-11-20 大连理工大学 The image search method of insertion is kept based on multi-angle of view Partial Reconstruction

Also Published As

Publication number Publication date
CN110727833A (en) 2020-01-24

Similar Documents

Publication Publication Date Title
CN106651519B (en) Personalized recommendation method and system based on label information
CN104881689B (en) A kind of multi-tag Active Learning sorting technique and system
CN106202256B (en) Web image retrieval method based on semantic propagation and mixed multi-instance learning
Zhang et al. Multilevel projections with adaptive neighbor graph for unsupervised multi-view feature selection
CN110059198A (en) A kind of discrete Hash search method across modal data kept based on similitude
CN106203483B (en) A kind of zero sample image classification method based on semantic related multi-modal mapping method
CN110532379B (en) Electronic information recommendation method based on LSTM (least Square TM) user comment sentiment analysis
CN106156333B (en) A kind of improvement list class collaborative filtering method of mosaic society's information
CN103559504A (en) Image target category identification method and device
CN110188228A (en) Cross-module state search method based on Sketch Searching threedimensional model
CN110348579A (en) A kind of domain-adaptive migration feature method and system
CN106951498A (en) Text clustering method
CN110110225B (en) Online education recommendation model based on user behavior data analysis and construction method
CN106250925B (en) A kind of zero Sample video classification method based on improved canonical correlation analysis
CN106844620B (en) View-based feature matching three-dimensional model retrieval method
CN110866134B (en) Image retrieval-oriented distribution consistency keeping metric learning method
CN109992676A (en) Across the media resource search method of one kind and searching system
CN106202999A (en) Microorganism high-pass sequencing data based on different scale tuple word frequency analyzes agreement
CN114943017A (en) Cross-modal retrieval method based on similarity zero sample hash
CN102930291B (en) Automatic K adjacent local search heredity clustering method for graphic image
Chen et al. RRGCCAN: Re-ranking via graph convolution channel attention network for person re-identification
Liu et al. Cross-Class Sample Synthesis for Zero-shot Learning.
CN114579794A (en) Multi-scale fusion landmark image retrieval method and system based on feature consistency suggestion
CN114896438A (en) Image-text retrieval method based on hierarchical alignment and generalized pooling graph attention machine mechanism
CN107392249A (en) A kind of density peak clustering method of k nearest neighbor similarity optimization

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant