CN108270608A - A kind of foundation of link prediction model and link prediction method - Google Patents
A kind of foundation of link prediction model and link prediction method Download PDFInfo
- Publication number
- CN108270608A CN108270608A CN201710004638.6A CN201710004638A CN108270608A CN 108270608 A CN108270608 A CN 108270608A CN 201710004638 A CN201710004638 A CN 201710004638A CN 108270608 A CN108270608 A CN 108270608A
- Authority
- CN
- China
- Prior art keywords
- model
- network
- link prediction
- data
- network data
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/01—Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Networks & Wireless Communication (AREA)
- Probability & Statistics with Applications (AREA)
- Signal Processing (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Telephonic Communication Services (AREA)
Abstract
The present invention provides a kind of method for building up of link prediction model, the link prediction model includes：Sequential is limited Boltzmann machine model and gradient promotes decision-tree model；The method includes：A large amount of network data is captured from internet or other multimedias, network data is pre-processed, network data is divided into historical data and available data, input timing is limited Boltzmann machine model, trains model parameter；The network topology characteristic of network data node pair is extracted, feature set is formed and input gradient promotes decision-tree model, train model parameter；The link prediction model is limited Boltzmann machine model including trained sequential and gradient promotes decision-tree model.Based on the link prediction model that this method is established, the present invention also provides a kind of link prediction method, this method can predict the all-links of network NextState.
Description
Technical field
The present invention relates to Internet technical fields, and in particular to a kind of foundation of link prediction model and link prediction side
Method, this method utilize the topological characteristic of network and deep learning model, and link prediction is carried out to large scale network.
Background technology
Along with the rapid development of internet and mobile communication technology, the contact between people becomes increasingly closer.It is logical
Internet and communication network are crossed, constitutes a huge complex network between men.In network it is interpersonal it is interactive,
The various aspects in life have been dissolved into exchange and influence.It also gradually attracts attention to the research of community network, and as current
One of research hotspot of scientific domain.In society, many people wish the structure and variation by analyzing community network, find
Contact principle between nodes knows to hide rule and community network topological features under general phenomenon
Relationship between nodal community feature and network node behavior trend, and then find the differentiation essence of community network, utilize this
A little information help people that resource and information processing is more effectively configured, and instruct commodity production, human lives, population management, nature
Management, interpretation and the decision of planning etc..One important research point of wherein network node behavior trend is exactly link prediction.
Link prediction method is used for describing the development trend in network future, the connection prediction that can be refined between node；
Incomplete either hiding side can also be found out in existing imperfect network.Traditional link prediction method generally utilizes
Network topology characteristic and nodal community are predicted using the method for machine learning.However these methods are all from microcosmic angle
Degree with node to carrying out link prediction for object, is unfavorable for the evolutionary Modeling to network macrostructure, thus its prediction effect
There are certain bottlenecks.
Invention content
It is an object of the invention to overcome drawbacks described above existing for current link prediction method, it is proposed that one kind is based on depth
The link prediction method of study, this method are limited adjacency matrix of the Boltzmann machine model to macroscopical sequential lower network using sequential
It is modeled, then using trained model as generation model, macro-forecast is carried out to the network linking state of next sequential.
On the other hand, network part topological characteristic is extracted from microcosmic angle, using machine learning model, (gradient promotes decision tree
Learning model) prediction network structure linking status.Finally by the prediction result Weighted Fusion of the two, the final chain of network is obtained
Road prediction result.This method describes the evolution of network from two angles of both macro and micro of network, to generate deep learning model
Based on, fusion machine learning model improves link prediction performance.
To achieve these goals, the present invention provides a kind of link prediction method based on deep learning, the links
Prediction model includes：Sequential is limited Boltzmann machine model and gradient promotes decision-tree model；The method includes：From internet
Or a large amount of network data is captured in other multimedias, network data is pre-processed, network data is divided into history number
According to and available data, input timing be limited Boltzmann machine model, train model parameter；Extract the net of network data node pair
Network topological characteristic, forms feature set and input gradient promotes decision-tree model, trains model parameter；The link prediction model
Boltzmann machine model is limited including trained sequential and gradient promotes decision-tree model.
In above-mentioned technical proposal, the method specifically includes：
Step S1) from internet or other multimedias a large amount of network data is captured, network data is pre-processed,
Network data is made not include isolated node or node pair；
Step S2) by certain time length network data is divided into timeslice, tectonic network figure G=under each timeslice
{GK,GK-1,…,G1, G is expressed as A={ A with sequential adjacency matrixK,AK-1,…,A1, then acknowledging time window is N, N<K,
Wherein { AN,AN-1,…,A2For historical data, { A1It is available data；
Step S3) historical data and available data input timing be limited Boltzmann machine model, train model parameter；
Step S4) by { GK,…,G2It is merged into basic network G '；With G1For standard set, select from G ' and jumped at a distance of for one
Node pair, form positive negative sample；And make positive and negative sample size consistent；The network topology characteristic of node pair is extracted, forms feature
Collect and input gradient promotes decision-tree model, train model parameter；
Step S5) the link prediction model training finishes, and the link prediction model is limited including trained sequential
Boltzmann machine model and gradient promote decision-tree model.
In above-mentioned technical proposal, the step S4) network topology characteristic include feature based on neighbours and based on network
The feature of migration.
Based on the link prediction model that the above method is established, the present invention also provides a kind of link prediction method, the sides
Method includes：
Step T1) crawl network data to be predicted, and pre-processed, network data is made not include isolated node or section
Point pair；
Step T2) by certain time length network data to be predicted is divided into timeslice, it is constructed under each timeslice
Network G={ GK,GK-1,…,G1,G0, wherein { GN,GN-1,…,G2For web-based history figure, { G1For existing network figure, { G0}
For network to be predicted；G is expressed as A={ A with sequential adjacency matrixK,AK-1,…,A1,A0}；Time window is N, N<K, when
Between window be moved along a unit, historical data becomes { AN-1,AN-2,…,A1, data two-value random initializtion to be predicted is
{A0, { A0It is available data, input timing is limited Boltzmann machine model, prediction result R1 is obtained after successive ignition；
Step T3) use { GK,GK-1,…,G1Construction basic network, utilize step S4) extraction feature set, input gradient promotion
Decision-tree model predicts { G0Under node connection status, obtain prediction result R2；
Step T4) weighting amalgamation result R1, R2, finally obtain the prediction result R after fusion.
In above-mentioned technical proposal, the step T4) realization process be：
If there is common node pair in R1 and R2, weight merging and obtain R=α R1+ (1- α) R2, α values exist
Between 0.5-0.7；If there are the node pair not having in R2, prediction result R=R1 in R1.
The advantage of the invention is that：
1st, link prediction method of the invention has merged deep learning and machine learning method, and network is described from two angles
Variation, overcome the deficiency of single model；And prediction be network NextState all-links, thus prediction effect is more
Comprehensively, more accurately；
2nd, link prediction method of the invention not only has heterogeneous networks a universality, but also to heterogeneous networks characteristic, no
Network with size has good robustness.
Description of the drawings
Fig. 1 is the sequence diagram of the link prediction method of the present invention.
Specific embodiment
The present invention will be further described in detail in the following with reference to the drawings and specific embodiments.
A kind of method for building up of link prediction model, the link prediction model include：Sequential is limited Boltzmann machine
(Temporal Restricted Boltzmann Machine, TRBM) model and gradient promote decision tree (Gradient
Boosting Decision Trees, GBDT) model；It the described method comprises the following steps：
Step S1) from internet or other multimedias a large amount of network data is captured, network data is pre-processed；
The temporal information on side is included in the network data；If the network data captured is not comprising isolated node or section
The network data of point pair, then can be used directly, otherwise need to pre-process the network data captured, delete isolated node
With node pair；
Step S2) by certain time length network data is divided into timeslice (snapshot), structure under each timeslice
Make network G={ GK,GK-1,…,G1, G is expressed as A={ A with sequential adjacency matrixK,AK-1,…,A1, when then confirming
Between window be N (N<K), wherein { AN,AN-1,…,A2For historical data, { A1It is available data；
Step S3) historical data and available data inputted into TRBM models, training pattern parameter；
Step S4) by { GK,…,G2It is merged into basic network G '；With G1For standard set, select from G ' and jumped at a distance of for one
Node pair, form positive negative sample；It due to positive and negative imbalanced training sets, needs to node to sampling so that positive and negative sample number
Amount is consistent；The network topology characteristic of node pair is extracted, form feature set and inputs GBDT model trainings, training pattern parameter；
Network topology characteristic includes the feature based on neighbours and the feature based on network wandering in the present embodiment；In this reality
It applies in example, neighbors feature Adamic-Adar；Migration is characterized as RootedPagerank.
Step S5) the link prediction model training finishes, and the link prediction model includes trained TRBM models
With GBDT models.
As shown in Figure 1, based on the link prediction model that the above method is established, the present invention also provides a kind of link prediction sides
Method, the method includes：
Step T1) crawl network data to be predicted, and pre-processed, network data is made not include isolated node or section
Point pair；
Step T2) by certain time length network data to be predicted is divided into timeslice, it is constructed under each timeslice
Network G={ GK,GK-1,…,G1,G0, wherein { GN,GN-1,…,G2For web-based history figure, { G1For existing network figure, { G0}
For network to be predicted；G is expressed as A={ A with sequential adjacency matrixK,AK-1,…,A1,A0}；Time window is N (N<K), when
Between window be moved along a unit, historical data becomes { AN-1,AN-2,…,A1, data two-value random initializtion to be predicted is
{A0, { A0It is available data, TRBM models are inputted, prediction result R1 is obtained after successive ignition；
Step T3) use { GK,GK-1,…,G1Construction basic network, utilize step S4) extraction feature set, input GBDT moulds
Type predicts { G0Under node connection status, obtain prediction result R2；
Step T4) weighting amalgamation result R1, R2, finally obtain the prediction result R after fusion.
If there is common node pair in R1 and R2, weighting merges R=α R1+ (1- α) R2, and α values are in 0.5-0.7
Between；If there are the node pair not having in R2 in R1, using the result of R1 as final result R=R1.
Claims (5)
1. a kind of method for building up of link prediction model, the link prediction model include：Sequential is limited Boltzmann machine model
Decision-tree model is promoted with gradient；The method includes：A large amount of network data is captured from internet or other multimedias, it is right
Network data is pre-processed, and network data is divided into historical data and available data, input timing is limited Boltzmann machine
Model trains model parameter；The network topology characteristic of network data node pair is extracted, feature set is formed and input gradient is promoted
Decision-tree model trains model parameter；The link prediction model is limited Boltzmann machine model including trained sequential
Decision-tree model is promoted with gradient.
2. the method for building up of link prediction model according to claim 1, which is characterized in that the method specifically includes：
Step S1) from internet or other multimedias a large amount of network data is captured, network data is pre-processed, makes net
Network data do not include isolated node or node pair；
Step S2) by certain time length network data is divided into timeslice, tectonic network figure G={ G under each timesliceK,
GK-1,…,G1, G is expressed as A={ A with sequential adjacency matrixK,AK-1,…,A1, then acknowledging time window is N, N<K, wherein
{AN,AN-1,…,A2For historical data, { A1It is available data；
Step S3) historical data and available data input timing be limited Boltzmann machine model, train model parameter；
Step S4) by { GK,…,G2It is merged into basic network G '；With G1For standard set, selected from G ' at a distance of the section jumped for one
Point pair, forms positive negative sample；And make positive and negative sample size consistent；The network topology characteristic of node pair is extracted, forms feature set simultaneously
Input gradient promotes decision-tree model, trains model parameter；
Step S5) the link prediction model training finishes, and the link prediction model is limited Bohr including trained sequential
Hereby graceful machine model and gradient promote decision-tree model.
3. the method for building up of link prediction model according to claim 2, which is characterized in that the step S4) network
Topological characteristic includes the feature based on neighbours and the feature based on network wandering.
4. a kind of link prediction method, based on the link prediction model realization that the method described in one of claim 2-3 is established, institute
The method of stating includes：
Step T1) crawl network data to be predicted, and pre-processed, network data is made not include isolated node or node
It is right；
Step T2) by certain time length network data to be predicted is divided into timeslice, tectonic network under each timeslice
Scheme G={ GK,GK-1,…,G1,G0, wherein { GN,GN-1,…,G2For web-based history figure, { G1For existing network figure, { G0To treat
The network of prediction；G is expressed as A={ A with sequential adjacency matrixK,AK-1,…,A1,A0}；Time window is N, N<K, time window
A unit is moved along, historical data becomes { AN-1,AN-2,…,A1, data two-value random initializtion to be predicted is { A0,
{A0It is available data, input timing is limited Boltzmann machine model, prediction result R1 is obtained after successive ignition；
Step T3) use { GK,GK-1,…,G1Construction basic network, utilize step S4) extraction feature set, input gradient promotion decision
Tree-model predicts { G0Under node connection status, obtain prediction result R2；
Step T4) weighting amalgamation result R1, R2, finally obtain the prediction result R after fusion.
5. link prediction method according to claim 1, which is characterized in that the step T4) realization process be：
If there is common node pair in R1 and R2, weight merging and obtain R=α R1+ (1- α) R2, α values are in 0.5-
Between 0.7；If there are the node pair not having in R2, prediction result R=R1 in R1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710004638.6A CN108270608B (en) | 2017-01-04 | 2017-01-04 | Link prediction model establishment and link prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710004638.6A CN108270608B (en) | 2017-01-04 | 2017-01-04 | Link prediction model establishment and link prediction method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108270608A true CN108270608A (en) | 2018-07-10 |
| CN108270608B CN108270608B (en) | 2020-04-03 |
Family
ID=62771669
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710004638.6A Active CN108270608B (en) | 2017-01-04 | 2017-01-04 | Link prediction model establishment and link prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108270608B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109639485A (en) * | 2018-12-13 | 2019-04-16 | 国家电网有限公司 | The monitoring method and device of electricity consumption acquisition communication link |
| CN110061961A (en) * | 2019-03-05 | 2019-07-26 | 中国科学院信息工程研究所 | A kind of anti-tracking network topological smart construction method and system based on limited Boltzmann machine |
| CN110445653A (en) * | 2019-08-12 | 2019-11-12 | 灵长智能科技(杭州)有限公司 | Network state prediction technique, device, equipment and medium |
| CN116132300A (en) * | 2022-09-15 | 2023-05-16 | 电子科技大学 | Link identification method based on gradient lifting decision tree feature combination |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103490948A (en) * | 2013-09-06 | 2014-01-01 | 华为技术有限公司 | Method and device for predicting network performance |
| US9129158B1 (en) * | 2012-03-05 | 2015-09-08 | Hrl Laboratories, Llc | Method and system for embedding visual intelligence |
-
2017
- 2017-01-04 CN CN201710004638.6A patent/CN108270608B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9129158B1 (en) * | 2012-03-05 | 2015-09-08 | Hrl Laboratories, Llc | Method and system for embedding visual intelligence |
| CN103490948A (en) * | 2013-09-06 | 2014-01-01 | 华为技术有限公司 | Method and device for predicting network performance |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109639485A (en) * | 2018-12-13 | 2019-04-16 | 国家电网有限公司 | The monitoring method and device of electricity consumption acquisition communication link |
| CN110061961A (en) * | 2019-03-05 | 2019-07-26 | 中国科学院信息工程研究所 | A kind of anti-tracking network topological smart construction method and system based on limited Boltzmann machine |
| CN110061961B (en) * | 2019-03-05 | 2020-08-25 | 中国科学院信息工程研究所 | Anti-tracking network topology intelligent construction method and system based on limited Boltzmann machine |
| CN110445653A (en) * | 2019-08-12 | 2019-11-12 | 灵长智能科技(杭州)有限公司 | Network state prediction technique, device, equipment and medium |
| CN116132300A (en) * | 2022-09-15 | 2023-05-16 | 电子科技大学 | Link identification method based on gradient lifting decision tree feature combination |
| CN116132300B (en) * | 2022-09-15 | 2024-04-30 | 电子科技大学 | Link identification method based on gradient lifting decision tree feature combination |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108270608B (en) | 2020-04-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108270608A (en) | A kind of foundation of link prediction model and link prediction method | |
| Gatti et al. | Large-scale multi-agent-based modeling and simulation of microblogging-based online social network | |
| CN104809501B (en) | A kind of computer system based on class brain coprocessor | |
| CN112084373B (en) | Graph embedding-based multi-source heterogeneous network user alignment method | |
| CN114373101A (en) | Image classification method for neural network architecture search based on evolution strategy | |
| CN104348829A (en) | Network security situation sensing system and method | |
| Tehseen et al. | A framework for the prediction of earthquake using federated learning | |
| CN112990378B (en) | Scene recognition method and device based on artificial intelligence and electronic equipment | |
| CN104361462B (en) | Social network influence maximization approach based on cultural gene algorithm | |
| CN110362728A (en) | Information-pushing method, device, equipment and storage medium based on big data analysis | |
| CN107086925B (en) | Deep learning-based internet traffic big data analysis method | |
| CN111126578A (en) | Joint data processing method, device and system for model training | |
| Garcia-Magarino et al. | Survivability strategies for emerging wireless networks with data mining techniques: A case study with NetLogo and RapidMiner | |
| CN107644268B (en) | Open source software project incubation state prediction method based on multiple features | |
| Popa et al. | Neural networks for production curve pattern recognition applied to cyclic steam optimization in diatomite reservoirs | |
| CN112000793A (en) | Man-machine interaction oriented dialogue target planning method | |
| CN110020379A (en) | It is a kind of to be embedded in the link prediction method for indicating model based on depth dynamic network | |
| Lingyu et al. | SMAM: Detecting rumors from microblogs with stance mining assisting task | |
| CN115658971A (en) | Attention mechanism-based multi-layer heterogeneous network node importance degree evaluation method | |
| Esmaili et al. | Effective synthetic data generation for fake user detection | |
| Ren et al. | [Retracted] A Study on Information Classification and Storage in Cloud Computing Data Centers Based on Group Collaborative Intelligent Clustering | |
| CN113743605A (en) | Method for searching smoke and fire detection network architecture based on evolution method | |
| Oghenekaro et al. | A hierarchical temporal memory based framework for mining event logs | |
| Jena et al. | A Comprehensive Study on Metaheuristics, Big Data and Deep Neural Network Strategies | |
| Kozlak et al. | Agent-based modelling of social organisations |
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 |