CN107784124A - A kind of LBSN super-networks link Forecasting Methodology based on time-space relationship - Google Patents
A kind of LBSN super-networks link Forecasting Methodology based on time-space relationship Download PDFInfo
- Publication number
- CN107784124A CN107784124A CN201711182961.9A CN201711182961A CN107784124A CN 107784124 A CN107784124 A CN 107784124A CN 201711182961 A CN201711182961 A CN 201711182961A CN 107784124 A CN107784124 A CN 107784124A
- Authority
- CN
- China
- Prior art keywords
- mrow
- user
- super
- msub
- node
- 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 39
- 238000013139 quantization Methods 0.000 claims abstract description 8
- 238000011478 gradient descent method Methods 0.000 claims abstract description 5
- 230000006399 behavior Effects 0.000 claims description 28
- 239000000203 mixture Substances 0.000 claims description 11
- 230000000694 effects Effects 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 6
- 230000007423 decrease Effects 0.000 claims description 5
- 230000003993 interaction Effects 0.000 claims description 4
- 230000003252 repetitive effect Effects 0.000 claims description 4
- 238000005259 measurement Methods 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 2
- 206010068052 Mosaicism Diseases 0.000 abstract description 3
- 238000009412 basement excavation Methods 0.000 abstract description 3
- 210000003765 sex chromosome Anatomy 0.000 abstract description 3
- 238000007418 data mining Methods 0.000 abstract description 2
- 230000036039 immunity Effects 0.000 abstract description 2
- 238000012549 training Methods 0.000 abstract description 2
- 238000011160 research Methods 0.000 description 4
- 230000002123 temporal effect Effects 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 235000007926 Craterellus fallax Nutrition 0.000 description 1
- 240000007175 Datura inoxia Species 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000036461 convulsion Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000005295 random walk Methods 0.000 description 1
- 230000000007 visual effect Effects 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
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9537—Spatial or temporal dependent retrieval, e.g. spatiotemporal queries
-
- 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
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- 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
- G06F16/95—Retrieval from the web
- G06F16/955—Retrieval from the web using information identifiers, e.g. uniform resource locators [URL]
- G06F16/9558—Details of hyperlinks; Management of linked annotations
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- General Health & Medical Sciences (AREA)
- Economics (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention relates to a kind of LBSN super-networks based on time-space relationship to link Forecasting Methodology, belongs to Data Mining.This method comprises the following steps：S1：Obtain data source；S2：Build supernetwork model；S3：Definition and quantization super-network side right value；S4：Based on the model, polytype super side structure of weighting is built first, by semantic relation different between different structure excavation users, finally by gradient descent method training pattern parameter, and then predicts the linking relationship in network.The present invention can effectively excavate a variety of incidence relations between node, can not only solve the sparse sex chromosome mosaicism in network by weighting super side structure, while can improve the noise immunity and stability of model, and its forecasting accuracy also has larger lifting.
Description
Technical field
The invention belongs to Data Mining, is related to a kind of LBSN super-networks link Forecasting Methodology based on time-space relationship.
Background technology
With the continuous development of computer information technology and the rapid popularization of internet, online social platform has become people
Live in an indispensable part, people can be established the friend relation net of oneself by this platform, be carried out with good friend
Instant exchange and interdynamic, this largely facilitates the life of people, particularly in recent years, location-based social networks
The appearance of (Location-Based Social Network, abbreviation LBSN) cause some location-based services in a short time by
The high praise of a large number of users, obtains great success.In LBSN, user can be registered in the position that he went, and
Share the place of registering of oneself to good friend, this behavior of registering can truly reflect the position activity of user so that virtual on line
Close ties are set up under the world and line between real world, new opportunities and challenges are brought for community network link prediction.
At this stage, link prediction can be largely classified into two methods：Method based on similitude and the method based on study.
The main thought of the wherein method based on similitude is the node pair without link for any two, calculates the phase between them
Like property fraction, fraction is higher, then more there may be link between them.Wherein representational method has common neighbours' index
(CN), preferential Link Indicator (PA), Adamic/Adar indexs (AA), Jaccard coefficients, Katz, Rooted PageRank etc.；
Method based on study is that link forecasting problem is converted into two classification problems, and the key of this method is that finding influence chain practices midwifery
Raw key feature, characteristic vector is established by these features, and carry out efficient model learning, it is accurately pre- so as to realize
Survey.
Isomery social networks is a kind of network comprising polytype node and side, and location-based social networks category
In a kind of isomery social networks, it is mainly reflected in it and possesses user node, nodes of locations, category node, user-position side, use
Family-user side etc..Most of link forecasting researches are concentrated mainly in homogeneous network at present, also with regard to there was only a type in network
Node (user node) and at (during user-user), the research for heterogeneous network is relatively fewer, therefore most of based on same
The link Forecasting Methodology of network forming network is no longer applicable.Predicted for heterogeneous network links, someone goes to study, examined from temporal information angle
Consider the correlation of the time or two users of link foundation in time-space relationship, research shows to consider that time factor can be effective
Improve the degree of accuracy of prediction；Also mode of the someone based on first path goes to study, and first path refers to connecting one of different nodes
Path, this paths have certain semantic information, such as user-comedy-user shows that two users like comedy, is based on
Different semantic paths, pass through the correlation between the mode calculate node such as random walk.Deta sparseness is this method solve to ask
Topic, but most of research is only for network is had no right, for weighted networks, it is still necessary to targetedly consider that network weight is special
Property；Supernetwork model can also solve the link forecasting problem in isomery social networks, and existing method is based on weighted supernetwork mould
Type, super triangular structure is weighted to predict the link in network by building, but the existing super triangular structure of weighting is only
Field node can be caught on being influenceed caused by link formation, and the super side structure of weighting for integrating other more horn of plenties, can not only
Enough alleviate Sparse sex chromosome mosaicism, while also help and further improve prediction accuracy.In addition, it is existing to be based on super-network
Method and temporal information could not be utilized, so its accuracy still has great room for promotion.
The content of the invention
In view of this, it is an object of the invention to provide a kind of LBSN super-networks based on time-space relationship to link prediction side
Method, the temporal and spatial correlations characteristic between isomerism and user for isomery social networks, it is proposed that a kind of " space-time-user-position
Put-classification " four layers of weighted supernetwork model, effectively temporal information has been dissolved into supernetwork model.In view of between user
Implicit behavior, potential site incidence relation and user preference the side right of weighted supernetwork is modified again, improve model
It is explanatory.Finally, based on revised weighted supernetwork model, define it is super while and it is super while structure, based on this excavate user it
Between incidence relation.
To reach above-mentioned purpose, the present invention provides following technical scheme：
A kind of LBSN super-networks link Forecasting Methodology based on time-space relationship, comprises the following steps：
S1：Obtain data source；Accurate data message with a high credibility is obtained from existing large-scale social network-i i-platform；Obtain
Data content include comment to position of friend relation, user between user, scoring and comment time, the longitude and latitude of position
Degree and the classification of position；
S2：Build supernetwork model；Including structure space-time subnet, social subnet, subnet of place and classification subnet, wherein when
Gap net is built-up to the time of registering of position using user, for excavating the space-time similitude between user；
S3：Definition and quantization super-network side right value；Believed by user force, hidden incidence relation, user preference, node degree
These four different modes are ceased to go to define the side right value in supernetwork model；
S4：By S1~S3 process, what a weighted supernetwork model is built, based on the model, is built first a variety of
The super side structure of weighting of type, by semantic relation different between different structure excavation users, finally by gradient decline side
Method training pattern parameter, and then predict the linking relationship in network, dead level, client layer, site layer and classification layer when being divided into.
Further, the step S2 is specially：
By primary data information (pdi), the friend relation list of user is extracted, register relation list and the position of user
Classification information；
S21：The time registered by user extracts space-time node；Space-time node refers to if two or more use
Family accesses some position jointly in some specific period, then the position is defined as a space-time node；Space-time section
Interest preference of the point reflection user in special time ad-hoc location；
S22：Construct space-time-user-four layers of position-classification supernetwork model；It is divided into space-time subnet, social subnet, position
Subnet and classification subnet；Incidence relation between four straton nets can be accessed under some types for user because of the interest preference of itself
Some points of interest, and these points of interest are registered, commented on and scored, if user have in special time it is special emerging
Interesting preference, then these users can be got up by same space-time node contacts；So far, four stratons under location-based social networks
Net structure is completed.
Further, the step S3 is specially：
S31：Pass through weights between user force quantization user-user；In location-based social networks, each user
Influence power be different；User force is divided into influence power between user's individual influence power and user, and respectively by chasing after
With network and behavior is followed to measure；
Behavior is followed in definition：If user v is once registered in the place that its good friend u registered, then it is assumed that user v is produced
The behavior of following to user v has been given birth to, v to u directed edge can be produced accordingly；
Network G is followed in definitionf=(Vf,Ef)：Wherein GfRepresent the directed networkses formed by following behavior, VfExpression is followed
User in network, EfDirected edge caused by behavior is followed in expression；
S311：User's individual influence power Iu：For measure user because itself behavior is to shadow caused by other users in network
Ring；Different time sections are considered by way of dividing isochronous surfaceThe influence power of middle user, by each time
The behavior composition of following of user follows network accordingly in section, divides S isochronous surface, tsFor s-th of isochronous surface, user
Final individual influence power is contributed by the individual influence power in each timeslice, and from current time it is more remote when
Between piece its individual influence power decay it is more；
In view of the presence of isolated node in network, using LeaderRank Algorithm for Solving user's individual influence powers, iteration
Formula is：
Wherein NuUser u neighbor node is represented,Represent user v out-degree；At steady state, LeaderRank will
Ground Node fraction is evenly distributed to every other node, and the final score of node is expressed as：
Iu=Iu(td)+Ig(td)/N
Wherein Ig(td) it is the fractions of Ground Node at steady state, N is total number of users；
As time goes by, the influence power of user can successively decrease therewith, defining attenuation function is：
Wu(ti)=exp (- ln2 × (tc-ti)/tm)
Wherein tcRepresent current time, tiRepresent i-th of timeslice, tmRepresent the half-life period that influence power reduces；
User u is in current time individual influence power total value IuFor：
Wherein Iu(ti) represent tiIndividual timeslice user u individual influence power；
S312：Influence power between user：Influence power I between useriIt is big to user v influence power that (u, v) is used for measure user u
It is small, the behavior of following is considered as to influence power between the interaction between user and measure user；
Place ratio I is followed in propositionpWith follow the ratio I that registerscBoth measurement indexs：
Wherein, M (v, u) represents that user v follows counting with registering for user u, PositionuRepresent user u position of registering
Sum is put, K (v, u) represents that user v follows user u number of always registering, CheckinuRepresent user u total degree of registering；
User force I (u, v) is：
Based on user force, quantify user-user side right value, for node to u, v, if u is to v user force
Height, then its corresponding sides weights should also be as height, the side right value between user and user is quantified as：
Wherein w:(u, w) ∈ S represent neighbor nodes of the user u in social subnet, and I (u, v) represents that user is social with it
Influence power size between subnet neighbor node；
S32：Defined by hidden incidence relation and quantify position-position side right value and classification-classification side right value；
Define the side right value between the side right value and classification and classification between position and position：
Wherein geodist (p, p') represents the distance between position p and p', Max | Wp| it is associated number for two positions
Maximum, w (p, p') be position p and p' by the number of user-association,For degree of incidence threshold value；
Wherein | P (c, c') | the place number for belonging to c and classification c' simultaneously is represented, Max | Pc| represent to belong to type c simultaneously
With other certain type of maximums counted；
S33：Defined by user preference and quantify user-position side right value；In location-based social networks, user
Scoring attribute to position can intuitively reflect preference of the user to this position；The position high to user preference is more
High weights, pass through exponential function amendment user-position side right value：
Wherein r (u, p) is scorings of the user u at the p of position；
S34：Defined by node out-degree and quantify remaining side right value.
Further, the step S4 is specially：
S41：Define super side and super side right weight；
Define the super side of three types：
A kind of super side SEI：Refer to the super side for only including a kind of type node, a kind of special super side is belonged in super net；
The super side SE of two classesII：Refer to side of the node between adjacent two layers subnet to composition, be characterized in only heterogeneous comprising two kinds
Node；
The super side SE of three classesIII：Refer to the side that adjacent three stratons net is formed, be characterized in only including three kinds of heterogeneous nodes；
Super side right refers to weights possessed by every super side again, is calculated by the side right value included in super side；
S42：Hyperlink is predicted：Super side based on the three types defined, propose to weight super side structure, and pass through weighting
Super side structure solves the hyperlink forecasting problem between user and user；Excavated by constructing polytype super side structure between node
Implicit semantic relation；
S421：The super triangular structure of weighting, including single weight super triangular structure, double super triangular structures and greatly of weighting
The super triangular structure of power；
It is single to weight super triangular structure：Super triangular structure is weighted to count by space-time node and the list that user node is formed
The similitude between user node is calculated, expression two users are liked in the activity of same time identical position；The super side structure of definition is equal
For closed loop configuration, there is directionality；
Double super triangular structures of weighting：Refer to comprising the continuous two super triangular structures of weighting；
The big super triangular structure of weighting：Refer to the triangular structure being made up of two super sides of three classes；
S422：Weight hypermatrix structure：User node likes movable at two related space-time nodes, and its weights is pair
The product of side right value should be surpassed：
S423：The super mixed structure of weighting：Including weighting super mixing I structures, super mixing II structures are weighted；Definition is respectively：
The super mixing I structures of weighting：Mixing I structures refer to increases a super side of one kind and group on the basis of single triangular structure
Into structure；
The super mixing II structures of weighting：Mixing II structures refer to increase a super side of one kind and group on the basis of rectangular configuration
Into structure；
Level is deeper, and related link circuits are longer, and super side structure is abundanter；
The super side structure of weighting includes：The super triangular structure of weighting, weighting hypermatrix structure, weight hypermatrix structure and
The super mixed structure of weighting；Different structures is different to link predicted impact degree, therefore its similitude is expressed as：
S (u, v)=θ1WS1(u,v)+θ2WS2(u,v)+......+θ19WS19(u,v)
Wherein θiFor the weight of i-th of super side structure of weighting, train to obtain by gradient descent method；Parameter renewal process
For：
Wherein, θi-oldRepresent the weights before repetitive exercise, θi-newThe weights after repetitive exercise are represented, λ represents study step
Long, y represents to whether there is link between user；When the changing value of each parameter is both less than some threshold value, parameter renewal has restrained,
Obtain optimized parameter set θ+, finally utilize optimized parameter set θ+Linking relationship user is predicted, when y values take 1,
Think that the link between user is present, otherwise it is assumed that the link between user is not present, its definition is as follows：
The beneficial effects of the present invention are：The present invention can be excavated effectively more between node by weighting super side structure
Kind incidence relation, can not only solve the sparse sex chromosome mosaicism in network, while can improve the noise immunity and stability of model, and
And its forecasting accuracy also has larger lifting.
Brief description of the drawings
In order that the purpose of the present invention, technical scheme and beneficial effect are clearer, the present invention provides drawings described below and carried out
Explanation：
Fig. 1 is the overview flow chart of the present invention；
Fig. 2 is the supernetwork model based on time-space relationship in LBSN；
Fig. 3 is two layers of supernetwork model.
Fig. 4 is three layers of supernetwork model.
Embodiment
Below in conjunction with accompanying drawing, the preferred embodiments of the present invention are described in detail.
As shown in Figure 1 and Figure 2, including：Data module is obtained, builds supernetwork model, definition and quantization network edge weights,
The common four module of LBSN super-networks link Forecasting Methodology based on time-space relationship.
The detailed implementation process of the detailed description below present invention.
S1：Obtain data source.The data of acquisition are yelp data, and the data are the opening data set of yelp websites.Obtain
Data content mainly include friend relation between user, comment and scoring of the user to shop, the longitude and latitude in shop is affiliated
Classification etc..
S2：Build supernetwork model.Because in location-based social networks, the foundation of link receives factors
Influence, such as time factor, positional factor, social factor etc..The present invention will be many by building the method for supernetwork model
Factor is digested, and applied to link prediction.Be specifically divided into four layers, respectively when dead level, client layer, site layer and classification
Layer.Concrete scheme is as follows：
S21：Construct space-time node.Space-time node refers to if two or more users are when some is specific
Between section have accessed some position jointly, then the position is defined as a space-time node.As can be seen that space-time node reflects
Interest preference of the user in special time ad-hoc location, compared to two users accessed some position, this interest jointly
Preference can more reflect the similitude between user.
S22：Construct space-time-user-four layers of position-classification supernetwork model.It is broadly divided into space-time subnet, social subnet,
Subnet of place and classification subnet.Incidence relation between four straton nets, which can be summarized as user, to be visited because of the interest preference of itself
Some points of interest under some types are asked, and these points of interest are registered, commented on and scored, if user is when specific
Interior to have special interest preference, then these users can be got up by same space-time node contacts.So far, location-based social activity
Four straton nets structure under network is completed, followed by the side right value for defining and quantifying super-network.
S3：Definition and quantization super-network side right value.Due in four layers of supernetwork model there is the node of four types and
The side of ten types, the side wherein in subnet have four kinds, and the side between subnet has six kinds, and the present invention goes to define by different methods
Side right value in network, is specifically divided into four kinds：It is inclined based on user based on implicit associations relation between position based on user force
The good and out-degree based on node.It is specific as follows：
S31：Pass through weights between user force quantization user-user.In location-based social networks, each user
Influence power be different.If some good friend is extremely low to our influence power, then we are difficult by the good friend and other
People produces some behaviors with contacting.So it is to improve model to be defined by the influence power of user and quantify user-user side right value
One of explanatory feasible method.User force is divided into influence power between user's individual influence power and user by the present invention, and
Measured respectively by following network and following behavior.
Behavior is followed in definition.If user v is once registered in the place that its good friend u registered, it is considered that
User v generates the behavior of following to user v, can produce v to u directed edge accordingly.
Network G is followed in definitionf=(Vf,Ef).Wherein GfRepresent the directed networkses formed by following behavior, VfExpression is followed
User in network, EfDirected edge caused by behavior is followed in expression.
S311：User's individual influence power.User's individual influence power IuFor measure user because itself behavior is to its in network
Influenceed caused by his user, be a kind of measure of global angle.Due to individual influence power can dynamic change over time,
Some users initially may compare actively, and its behavior of registering generates many and follows side, forms large effect power, lives afterwards
Jerk declines, then its influence power can gradually decrease to a stationary value.Therefore, for the influence power of accurate measure user, we
It is contemplated that time factor.
The present invention goes to consider the influence power of user in different time sections by way of dividing isochronous surface, by each time
The behavior composition of following of user follows network accordingly in sectionS isochronous surface has been divided herein, has been used
The final individual influence power in family is contributed by the individual influence power in each timeslice, and more remote from current time
Its individual influence power decay of timeslice is more.
In view of the presence of isolated node in network, the present invention is influenceed using LeaderRank Algorithm for Solving user individual
Power.LeaderRank algorithms are solved in PageRank because isolated node sorts knot and caused by by being introduced into Ground Node
The problem of fruit is not unique, and the algorithm the convergence speed is fast, and anti-noise ability is strong, can be good at being applied to the inventive method.Should
The iterative formula of algorithm is described as follows：
WhereinRepresent user v out-degree.At steady state, LeaderRank is uniform by Ground Node fraction
Every other node is distributed to, therefore the final score of node can be expressed as：
Iu=Iu(td)+Ig(td)/N (2)
Wherein Ig(td) it is the fractions of Ground Node at steady state, N is total number of users.
As time goes by, the influence power of user can successively decrease therewith, so definition attenuation function is：
Wu(ti)=exp (- ln2 × (tc-ti)/tm) (3)
Wherein tcRepresent current time, tiRepresent i-th of timeslice, tmRepresent the half-life period that influence power reduces.
User u is in current time individual influence power total value IuFor：
Wherein Iu(ti) represent tiIndividual timeslice user u individual influence power.
S312：Influence power between user.Influence power I between useriIt is big to user v influence power that (u, v) is used for measure user u
It is small, it is a kind of measure at local visual angle.Under normal circumstances, interaction times are more between two users, then between them
Influence power can be bigger.The interaction that the behavior of following is considered as between user in the present invention simultaneously carrys out influence power between measure user with this.
Place ratio I is followed in propositionpWith follow the ratio I that registerscBoth measurement indexs：
Wherein, M (v, u) represents that user v follows counting with registering for user u, PositionuRepresent user u position of registering
Sum is put, K (v, u) represents that user v follows user u number of always registering, CheckinuRepresent user u total degree of registering.
Analyzed more than, user force I (u, v) is：
Based on user force, user-user side right value can be quantified, for node to u, v, if u is to v user's shadow
It is high to ring power, then its corresponding sides weights should also be as height, so the side right value between user and user is quantified as：
Wherein w:(u, w) ∈ S represent neighbor nodes of the user u in social subnet.I (u, v) represents that user is social with it
Influence power size between subnet neighbor node.
S32：Defined by hidden incidence relation and quantify position-position side right value and classification-classification side right value.If some
User certain two position of connected reference in regular hour threshold value, then the two positions there is certain hidden association and close
System, similarly if two classifications appear in multiple positions simultaneously, then also there is certain hidden association between the two classifications
Relation, for example, from data statistics it can be found that classification Festivals and Arts＆Entertainment frequently appear in it is multiple
In the category attribute of position, implicit shows certain correlation be present between the two classifications.Based on considerations above, under
Formula defines the side right value between side right value and classification and classification between position and position.
Wherein Max | Wp| the maximum of number is associated for two positions, w (p, p') is position p and p' by user-association
Number,For degree of incidence threshold value, can be adjusted according to network characteristic and experimental performance.
Wherein | P (c, c') | the place number for belonging to c and classification c' simultaneously is represented, Max | Pc| represent to belong to type c simultaneously
With other certain type of maximums counted.
S33：Defined by user preference and quantify user-position side right value.In location-based social networks, user
Scoring attribute to position can intuitively reflect preference of the user to this position.Such as user u1In p1,p2,p3
Three positions carried out scoring, and sets forth 5,3,1 score value, discounting for scoring category of the user to this position
Property, then to every user-position side assignment 1/3, but actually, this is inaccurate, because if user u1To p3Scoring
For 1 point, it is unsatisfied, this when to this place to show user, should increase u1-p1Side right value, reduce u1-p3's
Side right value.As can be seen that should give the high position of user preference higher weights from this example, in the present invention, pass through finger
Number function amendment user-position side right value：
Wherein r (u, p) is scorings of the user u at the p of position.
S34：Defined by node out-degree and quantify remaining side right value.
S4：LBSN super-networks link Forecasting Methodology based on time-space relationship.By S1~S3 process, one has been built
Weighted supernetwork model, based on the model, structure weights super side structure and carries out link prediction.
S41：Define super side and super side right weight.In LBSN supernetwork models, there is polytype super side, for example use
Formed between family node and nodes of locations when super for one, the side formed between user node and space-time node is also one super
Side, due to different super sides, its heterogeneous nodes number included is different, therefore, defines the super side of three types.
A kind of super side SEI.A kind of super when referring to a kind of only super comprising type node, it is special in super net to belong to one kind
Super side.For example, one of two user nodes compositions is super when being referred to as a kind of super, a kind of super side is indicated in same straton net
Incidence relation between node, such as social subnet, refer to the friend relation between user.
The super side SE of two classesII.The super node while between referring to adjacent two layers subnet of two classes to composition while, be characterized in only wrapping
Containing two kinds of heterogeneous nodes.Such as the super side formed between user and nodes of locations or user and space-time node, referred to as two classes
Super side.
The super side SE of three classesIII.Three classes it is super while refer to that adjacent three stratons net forms while, be characterized in only heterogeneous comprising three kinds
Node.For example, the super side that user, position and category node are formed, the super side of referred to as three classes.
Such as Fig. 3, shown in Fig. 4, Fig. 3 is two layers adjacent of subnet, wherein, (T1-T2) to constitute a super side of one kind (orange
Side), it is designated as SEI(T1-T2)。(U1-T1) to constitute two classes super in (during yellow), it is designated as SEII(U1-T1), (U3-T1) also structure
It is super in (while green) into two classes, it is designated as SEII(U3-T1).Fig. 4 is adjacent three-layer network, wherein (U1-P1-C1) composition
One three class are super in (while light blue), are designated as SEIII(U1-P1-C1), (U3-P3-C1) constitute the super side (navy blue of three classes
Side), it is designated as SEIII(U3-P3-C1)。
Super side right weight.Super side right refers to weights possessed by every super side again, can be by the side right value meter that is included in super side
Obtain.For example, the super side SE of two classes in Fig. 3II(U1-T1), its super side right weightThree in Fig. 4
The super side right weight of class
S41：Hyperlink is predicted.Super side based on the three types defined, propose to weight super side structure, and pass through weighting
Super side structure solves the hyperlink forecasting problem between user and user.In conventional method, mainly by weighting super triangle
The correlation degree that shape structure is come between calculate node, its main thought are super by two by the co-occurrence node between different super sides
Frontier juncture joins, so as to obtain the similitude that super three-legged structure is used between node metric.This method is applied to heterogeneous network, Neng Goujian
Single extra connection efficiently caught between two nodes, prediction accuracy is also improved while alleviating Sparse Problem.
But super-network can not only describe the association between isomorphism node, while the association between heterogeneous nodes can be described, therefore
The network layer of consideration is deeper, and association chain is longer, then can more reflect fine-grained recessive association between egress.The present invention passes through
Construct the implicit semantic relation between polytype super side structure excavation node.
S411：The super triangular structure of weighting.Super triangular structure is weighted including single, double super triangular structures and greatly of weighting
The super triangular structure of power.It is defined as follows：
It is single to weight super triangular structure., can be by space-time node T1 and user node U1 in Fig. 3, the list that U3 is formed adds
The super triangular structure of power calculates U1, similitude between U3, and the semantic information of the structure representation is that two users are liked when identical
Between identical position activity.If comprising U1, U3 single super triangular structure number of weighting is more, and weights are bigger, then it is assumed that it
Between similitude it is also bigger, more there may be link.The super triangular structure includes the super side SE of two two classesII(U1-T1)
And SEII(T1-U3), the weight of super triangular structure is corresponds to the product of super side right weight, so weights are：
It is emphasized that the super side structure that the present invention defines is closed loop configuration, there is directionality.SoHereinafter similarly.
Double super triangular structures of weighting.Double triangle refers to comprising the continuous two super triangular structures of weighting, for example, Fig. 3
Middle SEII(U1-T1) and SEII(T1-U2) constitute a super triangular structure of weighting, SEII(U2-T2) and SEII(T2-U3) also structure
Into a super triangular structure of weighting, the two, which weight super triangular structure, can be combined into a double super triangle knot of weighting
Structure, for measuring U1, the similarity between U3, the semantic information of the structure is all liked with user U2 identical for user U1 and U3
The activity of time identical position.This pair weights super triangular structure weights as corresponding two single super triangular structure weights of weighting
Product, so weights are：
The big super triangular structure of weighting.The big super triangular structure of weighting refers to the triangle knot being made up of two super sides of three classes
Structure.For example, in Fig. 4, the super side SE of navy blueIII(Ui-Pj-Ck) and light blue super side SEIII(Ui-Pj-Ck) just constitute one greatly
The super triangular structure of weighting, the semantic information of the structure have the hobby of identical category for two users.Its weights is two three
The product of the super side right value of class, so weights are：
S412：Weight hypermatrix structure.In figure 3, super side SEII(U1-T1), SEI(T1-T2), SEII(T2-U3) can be with structure
Into a weighting hypermatrix structure, contain U1 in the weighting hypermatrix structure, two nodes of U3, available for measuring U1, U3 it
Between similitude, the semantic information of the structure likes movable at two related space-time nodes for user U1 and U3, its weights
For the product of corresponding super side right value：
S413：The super mixed structure of weighting.Including weighting super mixing I structures, super mixing II structures are weighted.It is defined as follows：
The super mixing I structures of weighting：Mixing I structures refer on the basis of single triangular structure increase a super side of one kind and
The structure of composition.For example, by super side SE in Fig. 1II(U1-T1),SEII(T1-U2),SEI(U2-U3) composition structure belong to mixing I
Structure, the good friend U2 that the semantic information of the structure representation is U3 are liked with U1 in the activity of identical time identical position.It is weighed
It is worth for corresponding single product for weighting super triangular structure weights and a kind of super side right value：
The super mixing II structures of weighting：Mixing II structures refer to increase a super side of one kind and group on the basis of rectangular configuration
Into structure.For example, by super side SE in Fig. 1II(U1-T1), SEI(T1-T2), SEII(U2-T2), SEI(U2-U3) composition structure category
In mixing II structures.Its weights is corresponding weighting hypermatrix structure weights and the product of a kind of super side right value：
It can be seen that level is deeper, related link circuits are longer, and super side structure is abundanter.The present invention lists wherein 19 kinds effectively
The super side structure of weighting, as shown in table 1.
It was found from above-mentioned analysis, the different super side structures of weighting has different semantic informations, such as S2 structures embody
The implication of position entropy, position entropy is meant that registers jointly if two users had in the place that many people went,
It is difficult to predict between the two people friend relation be present, because this is likely to be a kind of coincidence, but if two users are frequent
Registered in the place that a few went, then show to there may be certain relation between them.So position
Pouplarity also have influence to link prediction, and this influence can effectively be captured by S2 structures.And S3 can
Excavate a kind of short-term interest of user, short-term interest herein be construed to user only may just have in some period it is emerging
Interest, for example, on every Fridays at night 7 points go to the cinema and see a film.This interest only occurs in the specific period, but can more embody
Go out the individual character of user.
Because different structures is different to link predicted impact degree, therefore its similitude can be expressed as：
S (u, v)=θ1WS1(u,v)+θ2WS2(u,v)+......+θ19WS19(u,v) (19)
Wherein λiFor the weight of i-th of super side structure of weighting, can train to obtain by gradient descent method.Parameter updates
Process is as follows：
Wherein, λ represents Learning Step, and y represents to whether there is link between user.When the changing value of each parameter is both less than certain
During individual threshold value, parameter renewal has restrained, and obtains optimized parameter set θ+.Finally utilize optimized parameter set θ+The chain user
The relation of connecing is predicted, when y values take 1, it is believed that the link between user is present, otherwise it is assumed that the link between user is not present, it is fixed
Adopted formula is as follows：
Time factor is dissolved into supernetwork model by the present invention by way of being introduced into space-time node, is then based on user
Influence power, hidden incidence relation, four layers of weighted supernetwork model of user preference and node degree information architecture, improve solving for model
The property released, finally by a variety of super side structures of weighting, the semantic relation between user and user is excavated, not only solves Deta sparseness
Problem, while also improve prediction accuracy.It is emphasized that the present invention is a kind of having for weighted network link prediction
Efficacious prescriptions method, can preferably solve the link forecasting problem in weighted network.
Finally illustrate, preferred embodiment above is merely illustrative of the technical solution of the present invention and unrestricted, although logical
Cross above preferred embodiment the present invention is described in detail, it is to be understood by those skilled in the art that can be
Various changes are made to it in form and in details, without departing from claims of the present invention limited range.
Claims (4)
- A kind of 1. LBSN super-networks link Forecasting Methodology based on time-space relationship, it is characterised in that：This method comprises the following steps：S1：Obtain data source；Accurate data message with a high credibility is obtained from existing large-scale social network-i i-platform；The number of acquisition According to content include the comment to position of friend relation, user, scoring and comment time between user, position longitude and latitude and The classification of position；S2：Build supernetwork model；Including structure space-time subnet, social subnet, subnet of place and classification subnet, wherein when gap Net is built-up to the time of registering of position using user, for excavating the space-time similitude between user；S3：Definition and quantization super-network side right value；By user force, hidden incidence relation, user preference, node degree information this Four kinds of different modes go to define the side right value in supernetwork model；S4：By S1~S3 process, what a weighted supernetwork model is built, based on the model, builds polytype first The super side structure of weighting, different semantic relations between user is excavated by different structure, is instructed finally by gradient descent method Practice model parameter, and then predict the linking relationship in network, dead level, client layer, site layer and classification layer when being divided into.
- A kind of 2. LBSN super-networks link Forecasting Methodology based on time-space relationship according to claim 1, it is characterised in that： The step S2 is specially：By primary data information (pdi), the friend relation list of user, the classification of register relation list and the position of user are extracted Information；S21：The time registered by user extracts space-time node；Space-time node refers to if two or more users exist Some specific period accesses some position jointly, then the position is defined as a space-time node；Space-time node is anti- Reflect interest preference of the user in special time ad-hoc location；S22：Construct space-time-user-four layers of position-classification supernetwork model；It is divided into space-time subnet, social subnet, subnet of place With classification subnet；Incidence relation between four straton nets can access some under some types for user because of the interest preference of itself Point of interest, and these points of interest are registered, commented on and scored, if user has special interest inclined in special time Good, then these users can be got up by same space-time node contacts；So far, four straton net structures under location-based social networks Build completion.
- A kind of 3. LBSN super-networks link Forecasting Methodology based on time-space relationship according to claim 1, it is characterised in that： The step S3 is specially：S31：Pass through weights between user force quantization user-user；In location-based social networks, the shadow of each user It is different to ring power；User force is divided into influence power between user's individual influence power and user, and respectively by following net Network and behavior is followed to measure；Behavior is followed in definition：If user v is once registered in the place that its good friend u registered, then it is assumed that user v is generated Behavior of following to user v, v to u directed edge can be produced accordingly；Network G is followed in definitionf=(Vf,Ef)：Wherein GfRepresent the directed networkses formed by following behavior, VfNetwork is followed in expression In user, EfDirected edge caused by behavior is followed in expression；S311：User's individual influence power Iu：For measure user because itself behavior is on influence caused by other users in network；It is logical The mode for crossing division isochronous surface considers different time sectionsThe influence power of middle user, by each isochronous surface The behavior composition of following of middle user follows network accordingly, divides S isochronous surface, tsIt is final for s-th of isochronous surface, user Individual influence power contributed by the individual influence power in each timeslice, and the timeslice more remote from current time Its individual influence power decay is more；In view of the presence of isolated node in network, using LeaderRank Algorithm for Solving user's individual influence powers, iterative formula For：<mrow> <msub> <mi>I</mi> <mi>u</mi> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>+</mo> <mn>1</mn> <mo>)</mo> </mrow> <mo>=</mo> <munder> <mo>&Sigma;</mo> <mrow> <mi>v</mi> <mo>&Element;</mo> <mi>N</mi> <mrow> <mo>(</mo> <mi>u</mi> <mo>)</mo> </mrow> </mrow> </munder> <mfrac> <mn>1</mn> <msubsup> <mi>k</mi> <mi>v</mi> <mrow> <mi>o</mi> <mi>u</mi> <mi>t</mi> </mrow> </msubsup> </mfrac> <msub> <mi>I</mi> <mi>u</mi> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> </mrow>Wherein NuUser u neighbor node is represented,Represent user v out-degree；At steady state, LeaderRank will GroundNode fraction is evenly distributed to every other node, and the final score of node is expressed as：Iu=Iu(td)+Ig(td)/NWherein Ig(td) it is the fractions of GroundNode at steady state, N is total number of users；As time goes by, the influence power of user can successively decrease therewith, defining attenuation function is：Wu(ti)=exp (- ln2 × (tc-ti)/tm)Wherein tcRepresent current time, tiRepresent i-th of timeslice, tmRepresent the half-life period that influence power reduces；User u is in current time individual influence power total value IuFor：<mrow> <msub> <mi>I</mi> <mi>u</mi> </msub> <mo>=</mo> <munderover> <mo>&Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>0</mn> </mrow> <mi>k</mi> </munderover> <msub> <mi>I</mi> <mi>u</mi> </msub> <mrow> <mo>(</mo> <msub> <mi>t</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mo>&CenterDot;</mo> <msub> <mi>W</mi> <mi>u</mi> </msub> <mrow> <mo>(</mo> <msub> <mi>t</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> </mrow>Wherein Iu(ti) represent tiIndividual timeslice user u individual influence power；S312：Influence power between user：Influence power I between useri(u, v) is used for measure user u to user v influence power size, will chase after It is considered as influence power between the interaction between user and measure user with behavior；Place ratio I is followed in propositionpWith follow the ratio I that registerscBoth measurement indexs：<mrow> <msub> <mi>I</mi> <mi>p</mi> </msub> <mrow> <mo>(</mo> <mi>u</mi> <mo>,</mo> <mi>v</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <mi>M</mi> <mrow> <mo>(</mo> <mi>v</mi> <mo>,</mo> <mi>u</mi> <mo>)</mo> </mrow> </mrow> <mrow> <msub> <mi>Position</mi> <mi>u</mi> </msub> </mrow> </mfrac> </mrow><mrow> <msub> <mi>I</mi> <mi>c</mi> </msub> <mrow> <mo>(</mo> <mi>u</mi> <mo>,</mo> <mi>v</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <mi>K</mi> <mrow> <mo>(</mo> <mi>v</mi> <mo>,</mo> <mi>u</mi> <mo>)</mo> </mrow> </mrow> <mrow> <msub> <mi>Checkin</mi> <mi>u</mi> </msub> </mrow> </mfrac> </mrow>Wherein, M (v, u) represents that user v follows counting with registering for user u, PositionuRepresent that user u position of registering is total Number, K (v, u) represent that user v follows user u number of always registering, CheckinuRepresent user u total degree of registering；User force I (u, v) is：<mrow> <mi>I</mi> <mrow> <mo>(</mo> <mi>u</mi> <mo>,</mo> <mi>v</mi> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mo>&part;</mo> <mn>1</mn> </msub> <mo>&CenterDot;</mo> <msub> <mi>I</mi> <mi>u</mi> </msub> <mo>+</mo> <msub> <mo>&part;</mo> <mn>2</mn> </msub> <mo>&CenterDot;</mo> <msub> <mi>I</mi> <mi>p</mi> </msub> <mrow> <mo>(</mo> <mi>u</mi> <mo>,</mo> <mi>v</mi> <mo>)</mo> </mrow> <mo>+</mo> <msub> <mo>&part;</mo> <mn>3</mn> </msub> <mo>&CenterDot;</mo> <msub> <mi>I</mi> <mi>c</mi> </msub> <mrow> <mo>(</mo> <mi>u</mi> <mo>,</mo> <mi>v</mi> <mo>)</mo> </mrow> </mrow>Based on user force, quantify user-user side right value, for node to u, v, if u is high to v user force, Then its corresponding sides weights should also be as height, and the side right value between user and user is quantified as：<mrow> <mi>w</mi> <mrow> <mo>(</mo> <mi>u</mi> <mo>,</mo> <mi>v</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <mi>I</mi> <mrow> <mo>(</mo> <mi>u</mi> <mo>,</mo> <mi>v</mi> <mo>)</mo> </mrow> </mrow> <mrow> <munder> <mo>&Sigma;</mo> <mrow> <mi>w</mi> <mo>:</mo> <mrow> <mo>(</mo> <mi>u</mi> <mo>,</mo> <mi>w</mi> <mo>)</mo> </mrow> <mo>&Element;</mo> <mi>S</mi> </mrow> </munder> <mi>I</mi> <mrow> <mo>(</mo> <mi>u</mi> <mo>,</mo> <mi>w</mi> <mo>)</mo> </mrow> </mrow> </mfrac> </mrow>Wherein w:(u, w) ∈ S represent neighbor nodes of the user u in social subnet, and I (u, v) represents user and its social subnet Influence power size between neighbor node；S32：Defined by hidden incidence relation and quantify position-position side right value and classification-classification side right value；Define the side right value between the side right value and classification and classification between position and position：Wherein geodist (p, p') represents the distance between position p and p', Max | Wp| the maximum of number is associated for two positions Value, w (p, p') be position p and p' by the number of user-association,For degree of incidence threshold value；<mrow> <mi>w</mi> <mrow> <mo>(</mo> <mi>c</mi> <mo>,</mo> <msup> <mi>c</mi> <mo>&prime;</mo> </msup> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <mo>|</mo> <mi>P</mi> <mrow> <mo>(</mo> <mi>c</mi> <mo>,</mo> <msup> <mi>c</mi> <mo>&prime;</mo> </msup> <mo>)</mo> </mrow> <mo>|</mo> </mrow> <mrow> <mi>M</mi> <mi>a</mi> <mi>x</mi> <mo>|</mo> <msub> <mi>P</mi> <mi>c</mi> </msub> <mo>|</mo> </mrow> </mfrac> </mrow>Wherein | P (c, c') | the place number for belonging to c and classification c' simultaneously is represented, Max | Pc| represent to belong to type c and its simultaneously The maximum that he counts certain type ofly；S33：Defined by user preference and quantify user-position side right value；In location-based social networks, user's contraposition The scoring attribute put can intuitively reflect preference of the user to this position；The position high to user preference is higher Weights, pass through exponential function amendment user-position side right value：<mrow> <mi>w</mi> <mrow> <mo>(</mo> <mi>u</mi> <mo>,</mo> <mi>p</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <msup> <mi>e</mi> <mrow> <mi>r</mi> <mrow> <mo>(</mo> <mi>u</mi> <mo>,</mo> <mi>p</mi> <mo>)</mo> </mrow> </mrow> </msup> <mrow> <munder> <mo>&Sigma;</mo> <mrow> <mi>w</mi> <mo>:</mo> <mrow> <mo>(</mo> <mi>u</mi> <mo>,</mo> <mi>w</mi> <mo>)</mo> </mrow> <mo>&Element;</mo> <msub> <mi>P</mi> <mi>u</mi> </msub> </mrow> </munder> <msup> <mi>e</mi> <mrow> <mi>r</mi> <mrow> <mo>(</mo> <mi>u</mi> <mo>,</mo> <mi>w</mi> <mo>)</mo> </mrow> </mrow> </msup> </mrow> </mfrac> </mrow>Wherein r (u, p) is scorings of the user u at the p of position；S34：Defined by node out-degree and quantify remaining side right value.
- A kind of 4. LBSN super-networks link Forecasting Methodology based on time-space relationship according to claim 1, it is characterised in that： The step S4 is specially：S41：Define super side and super side right weight；Define the super side of three types：A kind of super side SEI：Refer to the super side for only including a kind of type node, a kind of special super side is belonged in super net；The super side SE of two classesII：Refer to side of the node between adjacent two layers subnet to composition, be characterized in only including two kinds of heterogeneous sections Point；The super side SE of three classesIII：Refer to the side that adjacent three stratons net is formed, be characterized in only including three kinds of heterogeneous nodes；Super side right refers to weights possessed by every super side again, is calculated by the side right value included in super side；S42：Hyperlink is predicted：Super side based on the three types defined, propose to weight super side structure, and by weighting super side Structure solves the hyperlink forecasting problem between user and user；By construct polytype super side structure excavate it is hidden between node Containing semantic relation；S421：The super triangular structure of weighting, including the super triangular structure of single weighting, the super triangular structure of double weightings and big weighting are super Triangular structure；It is single to weight super triangular structure：The list being made up of space-time node and user node weights super triangular structure to calculate use Similitude between the node of family, expression two users are liked in the activity of same time identical position；The super side structure of definition is to close Ring structure, there is directionality；Double super triangular structures of weighting：Refer to comprising the continuous two super triangular structures of weighting；The big super triangular structure of weighting：Refer to the triangular structure being made up of two super sides of three classes；S422：Weight hypermatrix structure：User node likes movable at two related space-time nodes, and its weights is corresponding super The product of side right value：S423：The super mixed structure of weighting：Including weighting super mixing I structures, super mixing II structures are weighted；Definition is respectively：The super mixing I structures of weighting：Mixing I structures refer to and increase a super side of one kind on the basis of single triangular structure and form Structure；The super mixing II structures of weighting：Mixing II structures refer to increase a super side of one kind on the basis of rectangular configuration and form Structure；Level is deeper, and related link circuits are longer, and super side structure is abundanter；The super side structure of weighting includes：The super triangular structure of weighting, weighting hypermatrix structure, weighting hypermatrix structure and weighting Super mixed structure；Different structures is different to link predicted impact degree, therefore its similitude is expressed as：S (u, v)=θ1WS1(u,v)+θ2WS2(u,v)+......+θ19WS19(u,v)Wherein θiFor the weight of i-th of super side structure of weighting, train to obtain by gradient descent method；Parameter renewal process is：<mrow> <msub> <mi>&theta;</mi> <mrow> <mi>i</mi> <mo>-</mo> <mi>n</mi> <mi>e</mi> <mi>w</mi> </mrow> </msub> <mo>=</mo> <msub> <mi>&theta;</mi> <mrow> <mi>i</mi> <mo>-</mo> <mi>o</mi> <mi>l</mi> <mi>d</mi> </mrow> </msub> <mo>+</mo> <mi>&lambda;</mi> <mrow> <mo>(</mo> <mi>y</mi> <mo>-</mo> <munderover> <mo>&Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mn>19</mn> </munderover> <msub> <mi>&theta;</mi> <mi>i</mi> </msub> <msub> <mi>W</mi> <mrow> <mi>S</mi> <mi>i</mi> </mrow> </msub> <mo>)</mo> </mrow> <mo>&times;</mo> <msub> <mi>W</mi> <mrow> <mi>S</mi> <mi>i</mi> </mrow> </msub> </mrow>Wherein, θi-oldRepresent the weights before repetitive exercise, θi-newThe weights after repetitive exercise are represented, λ represents Learning Step, y tables Show between user with the presence or absence of link；When the changing value of each parameter is both less than some threshold value, parameter renewal has restrained, and obtains most Excellent parameter sets θ+, finally utilize optimized parameter set θ+Linking relationship user is predicted, when y values take 1, it is believed that use Link between family is present, otherwise it is assumed that the link between user is not present, its definition is as follows：<mrow> <mi>P</mi> <mrow> <mo>(</mo> <mi>y</mi> <mo>|</mo> <mi>W</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <mi>exp</mi> <mrow> <mo>(</mo> <munderover> <mo>&Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mn>19</mn> </munderover> <msubsup> <mi>&theta;</mi> <mi>i</mi> <mo>+</mo> </msubsup> <msub> <mi>W</mi> <mrow> <mi>S</mi> <mi>i</mi> </mrow> </msub> <mo>)</mo> </mrow> </mrow> <mrow> <mn>1</mn> <mo>+</mo> <mi>exp</mi> <mrow> <mo>(</mo> <munderover> <mo>&Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mn>19</mn> </munderover> <msubsup> <mi>&theta;</mi> <mi>i</mi> <mo>+</mo> </msubsup> <msub> <mi>W</mi> <mrow> <mi>S</mi> <mi>i</mi> </mrow> </msub> <mo>)</mo> </mrow> </mrow> </mfrac> </mrow><mrow> <mi>y</mi> <mo>=</mo> <mfenced open = "{" close = ""> <mtable> <mtr> <mtd> <mn>1</mn> </mtd> <mtd> <mrow> <mi>P</mi> <mrow> <mo>(</mo> <mi>y</mi> <mo>|</mo> <mi>W</mi> <mo>)</mo> </mrow> <mo>&GreaterEqual;</mo> <mi>&xi;</mi> </mrow> </mtd> </mtr> <mtr> <mtd> <mn>0</mn> </mtd> <mtd> <mrow> <mi>o</mi> <mi>t</mi> <mi>h</mi> <mi>e</mi> <mi>r</mi> <mi>w</mi> <mi>i</mi> <mi>s</mi> <mi>e</mi> </mrow> </mtd> </mtr> </mtable> </mfenced> <mo>.</mo> </mrow>
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711182961.9A CN107784124B (en) | 2017-11-23 | 2017-11-23 | LBSN (location based service) hyper-network link prediction method based on space-time relationship |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711182961.9A CN107784124B (en) | 2017-11-23 | 2017-11-23 | LBSN (location based service) hyper-network link prediction method based on space-time relationship |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107784124A true CN107784124A (en) | 2018-03-09 |
| CN107784124B CN107784124B (en) | 2021-08-24 |
Family
ID=61430658
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711182961.9A Active CN107784124B (en) | 2017-11-23 | 2017-11-23 | LBSN (location based service) hyper-network link prediction method based on space-time relationship |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107784124B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109086373A (en) * | 2018-07-23 | 2018-12-25 | 东南大学 | A method of the fair link forecast assessment system of building |
| CN109635989A (en) * | 2018-08-30 | 2019-04-16 | 电子科技大学 | A kind of social networks link prediction method based on multi-source heterogeneous data fusion |
| CN110134883A (en) * | 2019-04-22 | 2019-08-16 | 哈尔滨英赛克信息技术有限公司 | A kind of isomery social network position entity anchor chain connects recognition methods |
| CN110851491A (en) * | 2019-10-17 | 2020-02-28 | 天津大学 | Network link prediction method based on multiple semantic influences of multiple neighbor nodes |
| CN111368788A (en) * | 2020-03-17 | 2020-07-03 | 北京迈格威科技有限公司 | Training method and device of image recognition model and electronic equipment |
| CN112765754A (en) * | 2020-12-31 | 2021-05-07 | 西安电子科技大学 | Time evolution graph design method based on super edge and suitable for satellite-ground network |
| CN113056753A (en) * | 2018-11-21 | 2021-06-29 | 贝宝公司 | Machine learning based on post-transaction data |
| CN113297500A (en) * | 2021-06-23 | 2021-08-24 | 哈尔滨工程大学 | Social network isolated node link prediction method |
| CN115242659A (en) * | 2022-08-09 | 2022-10-25 | 安徽大学 | High-order collective influence-based hyper-network node analysis method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103559407A (en) * | 2013-11-14 | 2014-02-05 | 北京航空航天大学深圳研究院 | Recommendation system and method for measuring node intimacy in weighted graph with direction |
| US20150271280A1 (en) * | 2012-11-21 | 2015-09-24 | Telefonaktiebolaget L M Ericsson (Publ) | Multi-objective server placement determination |
| CN107085616A (en) * | 2017-05-31 | 2017-08-22 | 东南大学 | A kind of false comment suspected sites detection method excavated based on multidimensional property in LBSN |
| CN107145527A (en) * | 2017-04-14 | 2017-09-08 | 东南大学 | Link prediction method based on first path in alignment isomery social networks |
-
2017
- 2017-11-23 CN CN201711182961.9A patent/CN107784124B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150271280A1 (en) * | 2012-11-21 | 2015-09-24 | Telefonaktiebolaget L M Ericsson (Publ) | Multi-objective server placement determination |
| CN103559407A (en) * | 2013-11-14 | 2014-02-05 | 北京航空航天大学深圳研究院 | Recommendation system and method for measuring node intimacy in weighted graph with direction |
| CN107145527A (en) * | 2017-04-14 | 2017-09-08 | 东南大学 | Link prediction method based on first path in alignment isomery social networks |
| CN107085616A (en) * | 2017-05-31 | 2017-08-22 | 东南大学 | A kind of false comment suspected sites detection method excavated based on multidimensional property in LBSN |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109086373A (en) * | 2018-07-23 | 2018-12-25 | 东南大学 | A method of the fair link forecast assessment system of building |
| CN109086373B (en) * | 2018-07-23 | 2021-01-12 | 东南大学 | Method for constructing fair link prediction evaluation system |
| CN109635989A (en) * | 2018-08-30 | 2019-04-16 | 电子科技大学 | A kind of social networks link prediction method based on multi-source heterogeneous data fusion |
| CN109635989B (en) * | 2018-08-30 | 2022-03-29 | 电子科技大学 | Social network link prediction method based on multi-source heterogeneous data fusion |
| CN113056753A (en) * | 2018-11-21 | 2021-06-29 | 贝宝公司 | Machine learning based on post-transaction data |
| CN110134883A (en) * | 2019-04-22 | 2019-08-16 | 哈尔滨英赛克信息技术有限公司 | A kind of isomery social network position entity anchor chain connects recognition methods |
| CN110134883B (en) * | 2019-04-22 | 2023-06-06 | 哈尔滨英赛克信息技术有限公司 | Heterogeneous social network location entity anchor link identification method |
| CN110851491A (en) * | 2019-10-17 | 2020-02-28 | 天津大学 | Network link prediction method based on multiple semantic influences of multiple neighbor nodes |
| CN110851491B (en) * | 2019-10-17 | 2023-06-30 | 天津大学 | Network link prediction method based on multiple semantic influence of multiple neighbor nodes |
| CN111368788A (en) * | 2020-03-17 | 2020-07-03 | 北京迈格威科技有限公司 | Training method and device of image recognition model and electronic equipment |
| CN111368788B (en) * | 2020-03-17 | 2023-10-27 | 北京迈格威科技有限公司 | Training method and device for image recognition model and electronic equipment |
| CN112765754A (en) * | 2020-12-31 | 2021-05-07 | 西安电子科技大学 | Time evolution graph design method based on super edge and suitable for satellite-ground network |
| CN112765754B (en) * | 2020-12-31 | 2023-12-22 | 西安电子科技大学 | Superside-based time evolution graph design method suitable for satellite-to-ground network |
| CN113297500A (en) * | 2021-06-23 | 2021-08-24 | 哈尔滨工程大学 | Social network isolated node link prediction method |
| CN115242659A (en) * | 2022-08-09 | 2022-10-25 | 安徽大学 | High-order collective influence-based hyper-network node analysis method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107784124B (en) | 2021-08-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107784124A (en) | A kind of LBSN super-networks link Forecasting Methodology based on time-space relationship | |
| CN105389332B (en) | It is a kind of geography social networks under user's similarity calculation method | |
| JP7468929B2 (en) | How to acquire geographical knowledge | |
| CN103064917B (en) | The high-impact customer group of a kind of specific tendency towards microblogging finds method | |
| Li et al. | A grouping method based on grid density and relationship for crowd evacuation simulation | |
| Zhao et al. | Impact of smart city planning and construction on economic and social benefits based on big data analysis | |
| CN106951524A (en) | Overlapping community discovery method based on node influence power | |
| CN103279887A (en) | Information-theory-based visual analysis method and system for micro-blog spreading | |
| CN107657043B (en) | Content-based mixed graph model image recommendation method | |
| Zhang et al. | Population evacuation analysis: considering dynamic population vulnerability distribution and disaster information dissemination | |
| CN110347897A (en) | Micro blog network emotion community detection method based on event detection | |
| CN106202488A (en) | Estimation user is to the method for physical event distance | |
| Stoyanov et al. | Internet of things water monitoring for a smart seaside city | |
| CN110008402A (en) | A kind of point of interest recommended method of the decentralization matrix decomposition based on social networks | |
| Srivastava et al. | Geo-social clustering of places from check-in data | |
| CN112016701B (en) | Abnormal change detection method and system integrating time sequence and attribute behaviors | |
| CN103179630A (en) | Data transmission method under opportunity network | |
| Yang et al. | Neighborhood coreness algorithm for identifying a set of influential spreaders in complex networks | |
| He et al. | Multi-view commercial hotness prediction using context-aware neural network ensemble | |
| Khatri et al. | Influence Maximization in social networks using discretized Harris’ Hawks Optimization algorithm | |
| CN109885644A (en) | A kind of importance appraisal procedure for Internet of Things Item Information searching order | |
| Zardi et al. | A multi-agent homophily-based approach for community detection in social networks | |
| Gao | Design and Implementation of 3D Animation Data Processing Development Platform Based on Artificial Intelligence | |
| Zhang et al. | Simulation of sports venue based on ant colony algorithm and artificial intelligence | |
| Xu et al. | Prediction of epidemic transmission path and risk management method in urban subway |
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 |