CN110136016A - A kind of multi-tag transmission method and system based on implicit association - Google Patents
A kind of multi-tag transmission method and system based on implicit association Download PDFInfo
- Publication number
- CN110136016A CN110136016A CN201910270828.1A CN201910270828A CN110136016A CN 110136016 A CN110136016 A CN 110136016A CN 201910270828 A CN201910270828 A CN 201910270828A CN 110136016 A CN110136016 A CN 110136016A
- Authority
- CN
- China
- Prior art keywords
- label
- node
- matrix
- information
- tag
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 68
- 230000005540 biological transmission Effects 0.000 title claims abstract description 21
- 239000011159 matrix material Substances 0.000 claims abstract description 56
- 238000012546 transfer Methods 0.000 claims abstract description 15
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000007781 pre-processing Methods 0.000 claims description 5
- 238000004458 analytical method Methods 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 5
- 238000003058 natural language processing Methods 0.000 description 4
- 230000006399 behavior Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000000644 propagated effect Effects 0.000 description 3
- 238000009412 basement excavation Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000007704 transition Effects 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 230000003466 anti-cipated effect Effects 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000002996 emotional effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 210000003813 thumb Anatomy 0.000 description 1
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
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2216/00—Indexing scheme relating to additional aspects of information retrieval not explicitly covered by G06F16/00 and subgroups
- G06F2216/03—Data mining
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- Economics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Transfer Between Computers (AREA)
Abstract
The present invention relates to a kind of multi-tag transmission method and system based on implicit association.This method comprises: the label information of part of node is it is known that and calculate probability transfer matrix to given network struction local network structure figure；The implicit association information between multi-tag is excavated based on label co-occurrence method；It generates node-label matrix and initializes the label information of unknown node；The label of each unknown node is updated according to probability transfer matrix and implicit association information；It is calculated based on node-label matrix and updates stop condition；Above-mentioned update step is executed repeatedly until meeting update stop condition or reaching given the number of iterations；The label information of the node of Unknown Label information in network is generated according to node-label matrix.The present invention can preferably excavate the incidence relation between label, the convergence rate of accelerated method, thus the label information of more acurrate comprehensive analysis catenet user.
Description
Technical field
The invention belongs to social networking application technical fields, and in particular to a kind of multi-tag propagation side based on implicit association
Method and system.
Background technique
In real world, there is universal connection and mutual dependence, the catenets of existing many between user
In, great deal of nodes label information missing, people can only be estimated by a small amount of node label information.With computer correlation skill
The development of art, more and more methods model complicated catenet, explore incidence relation between different nodes and
Potential rule, so that the Unknown Label information of node be better anticipated.Excavation is marked in the appearance of natural language processing (NLP) technology
Explicit associations relationship provides advantageous tool between label.
So far, it has been proposed that many multi-tag propagation algorithms.Label propagation algorithm is to be mentioned by Zhu in 2002 earliest
Out, basic thought is the label information for removing to predict unmarked node with the label information of marked node, is that multi-tag is propagated
The important milestone of algorithm.The advantage of the algorithm is that calculating process is simple, and method speed is fast, but the disadvantage is that algorithm stability
Poor, each result is quite different, meanwhile, relationship is intricate between real world label, but this method does not consider label
Between incidence relation, therefore, practice value it is not high.With the development of NLP technology, people excavate mark using NLP tool
Explicit information between label improves the accuracy of result, but in real world, and there are still unknown implicit passes between label
Connection, therefore, there are still biggish rooms for promotion for existing multi-tag transmission method.
Summary of the invention
The present invention is the incidence relation how further excavated between multi-tag for main technical problem, to predict
The Unknown Label information of node.The present invention provides a kind of multi-tag transmission method and system based on implicit association, into one
Step excavates the implicit association between multi-tag, is based on such implicit association, improves the performance of multi-tag transmission method.
To achieve the above object, technical solution provided by the invention is a kind of side that the multi-tag based on implicit association is propagated
Method the described method comprises the following steps:
Step A: local network structure figure is constructed to given network (such as catenet), is saved in the local network structure figure
The tag set of point is denoted as I, known to the label information of part of node (can be a small number of nodes)；It is general between calculate node
Rate transfer matrix T；
Step B: to given local network structure figure, it is based on label co-occurrence method, excavates the implicit pass between multi-tag
Join information；
Step C: the known label information based on part of nodes generates node-label matrix F, and initializes unknown node
Label information；
Step D: being updated the label of each unknown node, and updating principle is that the probability obtained according to step A turns
The implicit association information that matrix T and step B is obtained is moved to generate；
Step E: to the part of nodes after the primary update of the label of each unknown node, given based on step A
Label information more new node-label matrix F；
Step F: being based on node-label matrix F, calculates and updates stop condition；
Step G: executing step D-F repeatedly, until meeting update stop condition or reaching given the number of iterations, stops
It updates；
Step H: the label information of Unknown Label information node in network is generated according to node-label matrix F.
Further, above-mentioned steps B is specifically included, and excavates the implicit association between multi-tag based on label cooccurrence relation
Information, as the priori knowledge of the above method, priori knowledge P's is defined as:
Wherein, pijIndicate that the implicit association probability of label i and label j, t indicate label, li、ljIndicate i-th of tally set
Conjunction and j-th of tag set,M indicates number of labels, I=I1∪
I2∪I3∪ ..., tag set I is by cooccurrence relation I1, Cultural association I2, special event be associated with I3Equal labels composition.Different labels
Between implicit association probability constitute implicit association matrix P.The cooccurrence relation of label i and label j refer to the same node same time-division
With label i and label j；The Cultural association of label i and label j refer to due to factors such as culture, geographical locations, lead to some node
Label i and label j are distributed simultaneously；Label i is associated with the special event of label j to be referred to due to certain special in live network or again
The generation of major issue leads to some node while distributing label i and label j.
Further, above-mentioned steps C is specifically included, and for the node of part known label information, node i distributes label j
Probability be expressed as fij=1, the node unknown for other label informations, node i distributes the probability f of label jij=0.
Further, above-mentioned steps D is specifically included, and the unknown node label of label information is updated between given node
Probability transfer matrix T and multi-tag between implicit association information P, by t iteration, when the t+1 times iteration, node-mark
It is as follows to sign matrix F more new formula:
F(t+1)=λ TF(t)·P+(1-λ)F(0)
Wherein, F(t)Node-label matrix after indicating the t times iteration, λ are the hyper parameters of method.
Further, above-mentioned steps E, based on the label information F after given part labels information update iteration, acceleration side
Method convergence, rapid results.
Further, above-mentioned steps F updates stop condition and is defined as loss function loss less than some threshold given by man
Value, wherein loss function loss is specifically defined are as follows:
Wherein, F(t)Node-label matrix after indicating the t times iteration, relationship number in ξ=given localized network/true
Relationship number in network, represents the degree of rarefication of network.
Further, above-mentioned steps H, label generating mode are specifically defined are as follows: when node i distributes label j probability fijIt is greater than
When some threshold value (such as average level of the probability of each user distribution label), which is distributed into node.
The present invention also provides a kind of multi-tag broadcasting system based on implicit association, the system comprises: data acquisition module
Block, preprocessing module, priori knowledge building module, label propagation module and tag generation module；Wherein,
Data acquisition module is acquired for the interdependent node data to network, fetching portion node label information and
Relation data between node；
Preprocessing module constructs local area network according to source data information for reading the source data of data acquisition module acquisition
Network structure chart, the probability transfer matrix between calculate node, there is transition probability between the node of mutual-action behavior in a network is
1；Based on label information information initializing node-label matrix known to part, and initialize the label information of unknown node；
Priori knowledge constructs module, for being based on label co-occurrence method to given local network structure figure, excavates more marks
Implicit association information between label；
Label propagation module updates the label information of unknown node for iteration, until meeting stop condition or reaching repeatedly
Generation number generates node-label matrix, indicates the probability of node distribution label；
Tag generation module, node-label matrix result for being obtained according to label transmission method set each node point
Condition with label, meeting the condition is that node distributes the label, to generate the label information of unknown node.
Further, above-mentioned data acquisition module, in real world, since the factors such as privacy limit, specific acquisition section
Part relations data between the label information and node of point.
Further, the stop condition refers to: loss function loss is less than some threshold value given by man, wherein losing letter
Number loss is specifically defined are as follows:
Wherein, F(t)Node-label matrix after indicating the t times iteration, relationship number in ξ=given localized network/true
Relationship number in network represents the degree of rarefication of current given local network structure.
In conclusion the multi-tag transmission method and system provided by the invention based on implicit association, can be applied to microblogging
In all kinds of multi-tag classification problems such as Users' Interests Mining, have the advantage that
By the implicit association relationship between building label, the further relationship excavated between label, so that
Multi-tag transmission method more accurately, comprehensively positions the interest tags of user node.Meanwhile implicit association relationship between label
Introducing so that method in the present invention restrains more quickly compared to traditional multi-tag transmission method, algorithm complexity is low.
Detailed description of the invention
Fig. 1 is the implementation flow chart of the multi-tag transmission method provided by the invention based on implicit association.
Fig. 2 is the schematic diagram of one of applicating example provided by the invention.
Fig. 3 is the multi-tag broadcasting system structure chart provided by the invention based on implicit association.
Fig. 4 is label implicit association schematic diagram in embodiment.
Fig. 5 is the experimental result picture of the method for the present invention and other methods comparison in embodiment.
Fig. 6 is the number of iterations result figure of the method for the present invention and other methods comparison in embodiment.
Fig. 7 is the runing time result figure of the method for the present invention and other methods comparison in embodiment.
Specific embodiment
Below in conjunction with drawings and examples, the present invention is described in detail, it should be pointed out that described embodiment
It is intended merely to facilitate the understanding of the present invention, and does not play any restriction effect to it.
Fig. 1 is that the present invention provides a kind of implementation flow chart of multi-tag transmission method based on implicit association, such as Fig. 2 institute
To show, this method can be applied to the implicit interest digging analysis of social network user, specifically includes the following steps:
As in the present embodiment, by taking microblogging social networks as an example, using each user as a node in network, with
A line of the concern relation of user as network.Then, the network structure of a part is formd.It is a small number of in microblogging
The label informations of node users belongs to unknown message it is known that the label information of most of node users is hidden.
Step A: reading network data, constructs using social network user as node, and customer relationship is the social network diagram on side,
Probability transfer matrix T between calculate node；Specifically, probability transfer matrix calculating specifically includes following in the step A
Step:
Step A1: the similarity between calculate node user i and node users j, calculation method are as follows:
Wherein, IniAnd OutiRespectively indicate the out-degree and in-degree of node i.
Step A2: according to the similarity between node users described in step A1, calculating matrix W, calculation formula are as follows:
Step A3: diagonal matrix D is calculated according to matrix W described in step A2, wherein the diagonal element of diagonal matrix D is by W
Every row element is summed and is obtained, i.e. dii=∑J=1 ..., nwij, wherein n indicates all number of nodes；According to matrix D and W calculate node
Probability transfer matrix T between user, calculation formula are as follows:
T=D-1/2WD-1/2
Step B: known label information in this present embodiment is excavated between multi-tag using the method for label co-occurrence
Implicit association information.It specifically includes, the implicit association information between multi-tag is excavated based on label cooccurrence relation, as above-mentioned
The priori knowledge of method, priori knowledge P's is defined as:
Wherein, pijIndicate the implicit association probability of label i and label j,
I=I1∪I2∪I3∪ ..., tag set I is by cooccurrence relation I1, Cultural association I2, special event be associated with I3Equal labels composition.No
Implicit association matrix is constituted with the implicit association probability between label.
Step C: the known label information based on part of nodes user generates node-label matrix F, initializes unknown node
The label information of user, and initialize the number of iterations t=1.It specifically includes, for the node of part known label information, node i
The probability for distributing label j is fij=1, the node unknown for other label informations, node i distributes the probability f of label jij=0.
Specifically, in step C, a label value is distributed for each node, the node of Unknown Label is initial using 0
Change, part known node label is initialized based on true tag, row expression node, list indicating label, and node-label matrix is specific
It is defined as follows:
Step D: being updated the label of each unknown node user, update principle be obtained according to step A it is general
The implicit association information that rate transfer matrix T and step B is obtained.Specifically, to the label of label information unknown node user into
Row updates, according to the implicit association matrix P between probability transfer matrix T and multi-tag, by t iteration, the t+1 times iteration
When, node-label matrix F more new formula is as follows:
F(t+1)=λ TF(t)·P+(1-λ)F(0)
Wherein, F(t)Node-label matrix after indicating the t times iteration, λ are the hyper parameters of method.
Step E: after primary update, based on a small number of node users label informations more new node-label that step A is given
Matrix.
Step F: being based on node-label matrix, calculates and updates stop condition.Specifically, stop condition the following steps are included:
Step F1: calculating the loss function loss of current iteration, and calculation formula is as follows:
Step F2: according to loss described in step F1, judging whether it is less than some empirical value given by man, if so,
Then method reaches stop condition, otherwise, continues iteration.
Step G: executing step D-F repeatedly, until meeting update stop condition or reaching given the number of iterations, stops
It updates.
Step H: the label information of Unknown Label information node in network is generated according to node-label matrix.Specifically, section
Point user label information generate the following steps are included:
Step H1: the node generated according to above-mentioned steps G-label matrix calculates user and distributes the probability of label, and calculates
Each user distributes the average level of the probability of label；
Step H2: if the user of user node i distributes label probability and is greater than the average level, label j is distributed to
User node i (fij=1), otherwise label j is not assigned to user node i (fij=0).
As shown in figure 3, the present invention also provides a kind of multi-tag broadcasting system based on implicit association, specifically includes: data
Acquisition module, preprocessing module, priori knowledge building module, label propagation module and tag generation module.Wherein,
Data acquisition module 100 is acquired for the related data to the present embodiment microblog users, obtains the mark of user
The concern relation between information and user is signed, the label information of 10900 microblog users has been selected to form part in the present embodiment
Network structure, wherein known to 2200 user tag information；
Preprocessing module 200 constructs user according to source data information for reading the source data of data acquisition module acquisition
Between probability transfer matrix, there is between the user of mutual-action behavior transition probability in a network is 1, which includes closing
It infuses, forwards microblogging, thumbs up the behaviors such as microblogging, comment microblogging；Node-label square is initialized based on label information known to minority
Battle array, and initialize the label information of unknown node；
Priori knowledge constructs module 300, for the implicit association matrix between the information architecture label according to above-mentioned acquisition
Incidence relations are waited, further the implicit association between excavation label, in the present embodiment, co-occurrence number statistics such as figure between label
Shown in 4, for example, there are certain implicit associations for label " health " and label " tourism "；
Label propagation module 400 updates the label information of unknown node for iteration, until meeting stop condition or reaching
The number of iterations generates user-label matrix, and the stop condition refers to: loss function loss is less than some threshold value given by man,
Wherein loss function loss is specifically defined are as follows:
Wherein, F(t)Node-label matrix after indicating the t times iteration, relationship number in ξ=given localized network/true
Relationship number in network, represents the degree of rarefication of given local network structure；
Tag generation module 500, node-label matrix for being obtained according to label propagation module set each node distribution
The condition of label, meeting the condition is that node distributes the label, to generate the label information of unknown node.
To the convergence of the above method, it was demonstrated that as follows:
In given local network structure figure, node-label matrix F, the probability transfer matrix between node are initialized
T may be expressed as:
Wherein, subscript a indicates that the node of part known label information, subscript b indicate the node of Unknown Label information.This hair
The iterative formula of the multi-tag transmission method of bright design is as follows:
F(t+1)=λ TF(t)·P+(1-λ)F(0)
Matrix P wherein, 0≤pij≤1.FaNode-the label matrix for indicating known label information, is fixed and invariable.Cause
This, above formula can simplify are as follows: Fb←TbbFbP+TbaFaP, then:
Wherein,Indicate FbInitial value.Since T matrix is row regularization, and TbbIt is the submatrix of T, then:
It can release:
Due to (Tbb)n0 is converged to, thenTherefore, Fb=(I-Tbb)-1TbaFaIt can converge to constant
Value.
Fig. 5~Fig. 7 is the comparing result of the present invention and other methods in multiple performance indicators.Specifically, using this reality
The user part message in the live network that data acquisition module obtains in example is applied, user node is divided into 6 parts, data at random
Scale is ascending, is respectively labeled as 1#~6#.By method proposed by the present invention (hereinafter referred to as MLP-IA) and following side
Method compares experiment:
1) multi-tag transmission method (MLP): traditional multi-tag transmission method does not consider any priori knowledge.
2) the multi-tag transmission method (MLP-EA) based on explicit associations: on the basis of traditional multi-tag transmission method
On, introduce the explicit associations information (such as semantic association) between node；
3) it the multi-tag transmission method (MLP-EIA) based on explicit associations and implicit association: is propagated in traditional multi-tag
On the basis of method, the explicit associations information and implicit association information between node are introduced.
Experimental result is using rate of precision (Precision ratio), recall rate (Recall ratio) and F1 value (F1-
Measure) to control methods and the present invention relates to methods to assess.Comparing result Fig. 5 is shown, introduces more marks of implicit association
Label transmission method promoted in F1 value it is larger, to illustrate that present invention introduces the validity of the related information between label and can
Row.In compared with being introduced into the multi-tag transmission method of explicit information, the results showed that implicit information compared with explicit information more
Effectively.Because semantic information has more " noise " data in live network, have a certain impact to result.In addition, with
The Semantic of family text tends to diversification, emotional culture, has richer semantic information, is very difficult to capture user and really anticipates
Therefore figure introduces implicit association and compares explicit associations information, more effective for the promotion of methods and results.
In addition, method of the present invention can effectively accelerate the convergence of conventional method, reduce the number of iterations, method for improving speed
Degree, the present embodiment are based on identical data, and the number of iterations and runing time counted under distinct methods compares, as a result as schemed
Shown in 6 and Fig. 7.Compared with control methods, the method for introducing priori knowledge can greatly save the runing time of algorithm, accelerating algorithm
Convergence, quickly iteration to convergency value, especially on fairly large data set.
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also answered
Depending on protection scope of the present invention.
Claims (10)
1. a kind of multi-tag transmission method based on implicit association, which comprises the following steps:
Step A: to given network struction local network structure figure, the label information of part of node is it is known that and calculate section
Probability transfer matrix between point；
Step B: to the local network structure figure, the implicit association information between multi-tag is excavated based on label co-occurrence method；
Step C: the known label information based on the part of nodes generates node-label matrix, and initializes unknown node
Label information；
Step D: according to the probability transfer matrix and the implicit association information, the label of each unknown node is carried out
It updates；
Step E: to after the primary update of the label of each unknown node, the known label based on the part of nodes
Node-label matrix described in information update；
Step F: it is calculated based on the node-label matrix and updates stop condition；
Step G: executing step D-F repeatedly, until meeting the update stop condition or reaching given the number of iterations, stops
It updates；
Step H: the label information of the node of Unknown Label information in network is generated according to the node-label matrix.
2. the method as described in claim 1, which is characterized in that the calculation formula of probability transfer matrix described in step A are as follows:
T=D-1/2WD-1/2
Wherein, W is based on whether having Bian Xianglian and similarity calculation, calculation formula between node i and j are as follows:
Sim (i, j) indicates the similarity between node i and node j, calculation formula are as follows:
Wherein, IniAnd OutiRespectively indicate the out-degree and in-degree of node i；
D is a diagonal matrix, and diagonal element is obtained by the every row element summation of W.
3. the method as described in claim 1, which is characterized in that step B is excavated between multi-tag based on label cooccurrence relation
Implicit association information, as priori knowledge, the priori knowledge is defined as:
Wherein, pijIndicate the implicit association probability of label i and label j,I=
I1∪I2∪I3∪ ..., tag set I are by including cooccurrence relation I1, Cultural association I2, special event be associated with I3Set of tags inside
At；Implicit association probability between different labels constitutes implicit association matrix.
4. the method as described in claim 1, which is characterized in that in node-label matrix described in step C, row indicates node, column
Indicate label；For the node of part known label information, the probability of node i distribution label j is expressed as Fij=1, for other
The unknown node of label information, node i distribute the probability F of label jij=0.
5. the method as described in claim 1, which is characterized in that in step D, the unknown node label of label information be updated to
The probability transfer matrix T between node and the implicit association information P between multi-tag are determined, by t iteration, the t+1 times iteration
When, node-label matrix F more new formula is as follows:
F(t+1)=λ TF(t)·P+(1-λ)F(0)
Wherein, F(t)Node-label matrix after indicating the t times iteration, λ are hyper parameters.
6. the method as described in claim 1, which is characterized in that update stop condition described in step F is defined as loss function
Loss is less than some given threshold value, wherein loss function loss is defined as:
Wherein, F(t)Node-label matrix after indicating the t times iteration, relationship number/live network in ξ=given localized network
In relationship number, represent the degree of rarefication of network.
7. the method as described in claim 1, which is characterized in that the label generating mode in step H is defined as: when node i point
Probability f with label jijWhen greater than some threshold value, which is distributed into node.
8. the method for claim 7, which is characterized in that the threshold value is being averaged for the probability that each user distributes label
It is horizontal.
9. a kind of multi-tag broadcasting system based on implicit association characterized by comprising
Data acquisition module is acquired, fetching portion node label information and node for the interdependent node data to network
Between relation data；
Preprocessing module constructs local network structure figure, between calculate node for reading the data of data acquisition module acquisition
Probability transfer matrix, and node-label matrix is initialized based on label information known to part, and initialize unknown node
Label information；
Priori knowledge constructs module, for given local network structure figure, based on label co-occurrence method excavate multi-tag it
Between implicit association information；
Label propagation module updates the label information of unknown node for iteration, until meeting update stop condition or reaching repeatedly
Generation number generates node-label matrix, indicates the probability of node distribution label；
Tag generation module sets the condition of each node distribution label for node-label matrix, and meeting the condition is node
The label is distributed, to generate the label information of unknown node.
10. system as claimed in claim 9, which is characterized in that it is small that the update stop condition is defined as loss function loss
In some given threshold value, wherein loss function loss is defined as:
Wherein, F(t)Node-label matrix after indicating the t times iteration, relationship number/live network in ξ=given localized network
In relationship number, represent the degree of rarefication of network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910270828.1A CN110136016B (en) | 2019-04-04 | 2019-04-04 | Multi-label propagation method and system based on implicit association |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910270828.1A CN110136016B (en) | 2019-04-04 | 2019-04-04 | Multi-label propagation method and system based on implicit association |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110136016A true CN110136016A (en) | 2019-08-16 |
| CN110136016B CN110136016B (en) | 2021-06-29 |
Family
ID=67569396
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910270828.1A Active CN110136016B (en) | 2019-04-04 | 2019-04-04 | Multi-label propagation method and system based on implicit association |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110136016B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111160398A (en) * | 2019-12-06 | 2020-05-15 | 重庆邮电大学 | Missing label multi-label classification method based on example level and label level association |
| CN111582402A (en) * | 2020-05-15 | 2020-08-25 | 北京市商汤科技开发有限公司 | Target detection method and device, electronic equipment and storage medium |
| CN112579793A (en) * | 2020-12-24 | 2021-03-30 | 北京创鑫旅程网络技术有限公司 | Model training method and POI (Point of interest) label detection method and device |
| CN113763167A (en) * | 2021-08-11 | 2021-12-07 | 杭州盈火网络科技有限公司 | Blacklist mining method based on complex network |
| CN114398993A (en) * | 2022-01-18 | 2022-04-26 | 平安科技（深圳）有限公司 | Search information recall method, system, apparatus and medium based on tag data |
| CN114615090A (en) * | 2022-05-10 | 2022-06-10 | 富算科技(上海)有限公司 | Data processing method, system, device and medium based on cross-domain label propagation |
| WO2022193627A1 (en) * | 2021-03-15 | 2022-09-22 | 华南理工大学 | Markov chain model-based paper collective classification method and system, and medium |
| CN111582402B (en) * | 2020-05-15 | 2024-06-11 | 北京市商汤科技开发有限公司 | Target detection method and device, electronic equipment and storage medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102750375A (en) * | 2012-06-21 | 2012-10-24 | 武汉大学 | Service and tag recommendation method based on random walk |
| US8560958B1 (en) * | 2003-04-16 | 2013-10-15 | The Mathworks, Inc. | Signal navigation and label propagation in block diagrams |
| US20140067808A1 (en) * | 2012-09-06 | 2014-03-06 | International Business Machines Corporation | Distributed Scalable Clustering and Community Detection |
| CN103970863A (en) * | 2014-05-08 | 2014-08-06 | 清华大学 | Method and system for excavating interest of microblog users based on LDA theme model |
| CN105631157A (en) * | 2016-01-13 | 2016-06-01 | 西安电子科技大学 | Label propagation method based on propagation limitation |
| CN106789588A (en) * | 2016-12-30 | 2017-05-31 | 东软集团股份有限公司 | Label transmission method and device |
| CN108427730A (en) * | 2018-02-27 | 2018-08-21 | 江苏大学 | It is a kind of that method is recommended based on the Social Label of random walk and condition random field |
| CN108898506A (en) * | 2018-05-31 | 2018-11-27 | 西安理工大学 | A kind of overlapping corporations method for digging propagated based on multi-tag |
-
2019
- 2019-04-04 CN CN201910270828.1A patent/CN110136016B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8560958B1 (en) * | 2003-04-16 | 2013-10-15 | The Mathworks, Inc. | Signal navigation and label propagation in block diagrams |
| CN102750375A (en) * | 2012-06-21 | 2012-10-24 | 武汉大学 | Service and tag recommendation method based on random walk |
| US20140067808A1 (en) * | 2012-09-06 | 2014-03-06 | International Business Machines Corporation | Distributed Scalable Clustering and Community Detection |
| CN103970863A (en) * | 2014-05-08 | 2014-08-06 | 清华大学 | Method and system for excavating interest of microblog users based on LDA theme model |
| CN105631157A (en) * | 2016-01-13 | 2016-06-01 | 西安电子科技大学 | Label propagation method based on propagation limitation |
| CN106789588A (en) * | 2016-12-30 | 2017-05-31 | 东软集团股份有限公司 | Label transmission method and device |
| CN108427730A (en) * | 2018-02-27 | 2018-08-21 | 江苏大学 | It is a kind of that method is recommended based on the Social Label of random walk and condition random field |
| CN108898506A (en) * | 2018-05-31 | 2018-11-27 | 西安理工大学 | A kind of overlapping corporations method for digging propagated based on multi-tag |
Non-Patent Citations (4)
| Title |
|---|
| FATEMEH FARAHNAK-GHAZANI ET AL: ""Multi-label Classification with Feature-aware Implicit Encoding and Generalized Cross-entropy Loss "", 《2016 24TH IRANIAN CONFERENCE ON ELECTRICAL ENGINEERING (ICEE)》 * |
| 刘培奇 等: ""基于LDA 主题模型的标签传递算法"", 《计算机应用》 * |
| 王 帅 等: ""基于显式和隐式社交网络的混合推荐"", 《计算机应用与软件》 * |
| 葛艳艳: ""基于社会标签系统的推荐技术研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111160398A (en) * | 2019-12-06 | 2020-05-15 | 重庆邮电大学 | Missing label multi-label classification method based on example level and label level association |
| CN111160398B (en) * | 2019-12-06 | 2022-08-23 | 重庆邮电大学 | Missing label multi-label classification method based on example level and label level association |
| CN111582402A (en) * | 2020-05-15 | 2020-08-25 | 北京市商汤科技开发有限公司 | Target detection method and device, electronic equipment and storage medium |
| CN111582402B (en) * | 2020-05-15 | 2024-06-11 | 北京市商汤科技开发有限公司 | Target detection method and device, electronic equipment and storage medium |
| CN112579793A (en) * | 2020-12-24 | 2021-03-30 | 北京创鑫旅程网络技术有限公司 | Model training method and POI (Point of interest) label detection method and device |
| CN112579793B (en) * | 2020-12-24 | 2024-04-30 | 北京创鑫旅程网络技术有限公司 | Model training method, POI label detection method and device |
| WO2022193627A1 (en) * | 2021-03-15 | 2022-09-22 | 华南理工大学 | Markov chain model-based paper collective classification method and system, and medium |
| CN113763167A (en) * | 2021-08-11 | 2021-12-07 | 杭州盈火网络科技有限公司 | Blacklist mining method based on complex network |
| CN113763167B (en) * | 2021-08-11 | 2023-11-17 | 杭州盈火网络科技有限公司 | Blacklist mining method based on complex network |
| CN114398993A (en) * | 2022-01-18 | 2022-04-26 | 平安科技（深圳）有限公司 | Search information recall method, system, apparatus and medium based on tag data |
| CN114398993B (en) * | 2022-01-18 | 2024-05-14 | 平安科技（深圳）有限公司 | Search information recall method, system, device and medium based on tag data |
| CN114615090A (en) * | 2022-05-10 | 2022-06-10 | 富算科技(上海)有限公司 | Data processing method, system, device and medium based on cross-domain label propagation |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110136016B (en) | 2021-06-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110136016A (en) | A kind of multi-tag transmission method and system based on implicit association | |
| Chen et al. | Privacy preserving point-of-interest recommendation using decentralized matrix factorization | |
| CN113360915B (en) | Intelligent contract multi-vulnerability detection method and system based on source code diagram representation learning | |
| Heidari et al. | SMG: Fast scalable greedy algorithm for influence maximization in social networks | |
| WO2017211051A1 (en) | Mining method and server for social network account of target subject, and storage medium | |
| Mo et al. | Cross-task crowdsourcing | |
| JP7442696B2 (en) | An approach for differentially private federated learning based on voting | |
| CN112765653B (en) | Multi-source data fusion privacy protection method based on multi-privacy policy combination optimization | |
| CN107480213B (en) | Community detection and user relation prediction method based on time sequence text network | |
| CN110322003B (en) | Gradient-based graph confrontation sample generation method for document classification by adding false nodes | |
| Tan et al. | ECRModel: An elastic collision-based rumor-propagation model in online social networks | |
| CN112615888A (en) | Threat assessment method and device for network attack behavior | |
| CN111353554B (en) | Method and device for predicting missing user service attributes | |
| Buccafurri et al. | A new form of assortativity in online social networks | |
| Yu et al. | Fast budgeted influence maximization over multi-action event logs | |
| CN115409155A (en) | Information cascade prediction system and method based on Transformer enhanced Hooke process | |
| Liu et al. | Strong social graph based trust-oriented graph pattern matching with multiple constraints | |
| Gu et al. | LR-BA: Backdoor attack against vertical federated learning using local latent representations | |
| US10719779B1 (en) | System and means for generating synthetic social media data | |
| Williams et al. | Scaling POMDPs for dialog management with composite summary point-based value iteration (CSPBVI) | |
| CN110956553A (en) | Community structure division method based on social network node dual-label propagation algorithm | |
| Stoop et al. | Mesocopic comparison of complex networks based on periodic orbits | |
| CN109145633A (en) | Track data method for secret protection, electronic equipment, storage medium and system | |
| CN114092269A (en) | Time sequence data prediction method and device based on improved generalized network vector model | |
| Yang et al. | Information propagation dynamics model based on implicit cluster structure network |
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 |