CN108650614A - A kind of the location of mobile users prediction technique and device of automatic deduction social relationships - Google Patents
A kind of the location of mobile users prediction technique and device of automatic deduction social relationships Download PDFInfo
- Publication number
- CN108650614A CN108650614A CN201810222692.2A CN201810222692A CN108650614A CN 108650614 A CN108650614 A CN 108650614A CN 201810222692 A CN201810222692 A CN 201810222692A CN 108650614 A CN108650614 A CN 108650614A
- Authority
- CN
- China
- Prior art keywords
- user
- social relationships
- time
- social
- individual
- 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 134
- 238000012795 verification Methods 0.000 claims abstract description 18
- 239000011159 matrix material Substances 0.000 claims description 104
- 230000002452 interceptive effect Effects 0.000 claims description 57
- 230000002123 temporal effect Effects 0.000 claims description 54
- 230000003542 behavioural effect Effects 0.000 claims description 26
- 230000003993 interaction Effects 0.000 claims description 20
- 239000013598 vector Substances 0.000 claims description 18
- 230000008569 process Effects 0.000 claims description 11
- 238000001514 detection method Methods 0.000 claims description 7
- 238000010606 normalization Methods 0.000 claims description 7
- 230000000694 effects Effects 0.000 claims description 6
- 239000000284 extract Substances 0.000 claims description 4
- 238000009825 accumulation Methods 0.000 claims description 3
- 239000012141 concentrate Substances 0.000 claims description 3
- 230000001186 cumulative effect Effects 0.000 claims description 3
- 230000004927 fusion Effects 0.000 claims description 3
- 238000012417 linear regression Methods 0.000 claims description 3
- 230000011514 reflex Effects 0.000 claims description 2
- 125000006850 spacer group Chemical group 0.000 claims 1
- 239000000843 powder Substances 0.000 abstract description 2
- 230000006399 behavior Effects 0.000 description 81
- 238000010586 diagram Methods 0.000 description 13
- 230000008859 change Effects 0.000 description 6
- 238000013461 design Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 235000012054 meals Nutrition 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 230000000903 blocking effect Effects 0.000 description 1
- 238000000205 computational method Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 230000006806 disease prevention Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 238000013277 forecasting method Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 230000002045 lasting effect Effects 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 230000003442 weekly effect Effects 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/02—Services making use of location information
- H04W4/021—Services related to particular areas, e.g. point of interest [POI] services, venue services or geofences
-
- 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
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- 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
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/02—Services making use of location information
- H04W4/023—Services making use of location information using mutual or relative location information between multiple location based services [LBS] targets or of distance thresholds
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/02—Services making use of location information
- H04W4/029—Location-based management or tracking services
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Marketing (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Tourism & Hospitality (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Development Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention belongs to mobile behavior electric powder prediction, specially a kind of the location of mobile users prediction technique and device of automatic deduction social relationships.The present invention includes:The individual behavior record of user is obtained from user's mobile behavior log database;Social relationships type between deduction user accordingly;It is to connect side using social relationships type of the user between node, two users, builds user's social relation network;It is recorded using user's individual behavior, builds the chronological discrete motion track sequence of user;Social relationships subgraph is generated using the German number of outstanding person's card, builds zero model, the magnitude relationship of more each social relationships subgraph value of statistical indicant under live network and zero model determines user group's social relationships die body;Carry out the verification of user's individual social relationships die body;Markov forecast techniques device, acquaintance's fallout predictor, known stranger's fallout predictor and output controller, the Future Positions for predicting the user are established respectively.The accuracy of position prediction can be improved in the present invention, protects the individual privacy of user.
Description
Technical field
The invention belongs to mobile behavior electric powder predictions, and in particular to a kind of mobile subscriber of automatic deduction social relationships
Position predicting method and device.
Background technology
In recent years, the generation of extensive human behavior track data promotes academia to burst out large quantities of quarters with acquisition
Draw the innovation research of mankind's Move Mode.The proof analysis of related mankind's movement and scale-model investigation are different field in human lives
Practical application scene played huge effect, such as position prediction, disease prevention and control, traffic trip planning, data
Shared and disaster response etc..As one of extremely important application, the project of prediction user's Future Positions is known as academia
With the research hotspot of industrial quarters.
Currently, the method for relevant user position prediction can be mainly divided into two major classes:One kind is based on user itself history
The method of motion track, another kind of is then the method in conjunction with user's social relationships.
In those early years researchers devise a series of prediction techniques according to the historical movement path information of user itself, wherein
A kind of most representational method is the prediction technique based on Markov chain.Song in 2006 et al. is published in《IEEE
Transactions on Mobile Computing》On《Evaluating Next-Cell Predictors with
Extensive Wi-Fi Mobility Data》And Lu in 2013 et al. is published in《Scientific Reports》On
《Approaching the Limit of Predictability in Human Mobility》Equal work are described based on not
With the prediction technique of exponent number Markov chain, article is found by analysis compared to complicated high-order Markov forecast techniques method, letter
Single low order (single order or second order) Markov forecast techniques method can reach preferable predictablity rate.The advantage of the type method
Be to realize it is fairly simple, but predictablity rate still have it is to be hoisted.
With the rapid development of the fields such as the social networks of social networks, wireless communication and mobile computing in recent years, grind
Social relationships of the person of studying carefully gradually by user on social networks use the design of position predicting method.2015
Zhang et al. exists《IEEE Transactions on Computers》On《NextCell:Predicting Location
Using Social Interplay from Cell Phone Traces》And Jia in 2016 et al. exists《ACM
Transactions on Intelligent Systems and Technology》On deliver《Location
Prediction:A Temporal-Spatial Bayesian Model》Equal work utilize the social relationships such as the friend of user
Motion track predicts the Future Positions of the user.Such methods are made a general survey of, their general character is to be utilized the friend of user, same
Acquaintances' social relationships such as thing/classmate carry out the position prediction service for user, therefore compare first kind method, and such methods are being predicted
There is obvious promotion in accuracy rate.
However, remaining problem in existing Study on Forecasting Method, i.e., user's social relationships are considered incomplete
Face.The existing prediction technique using user's social relationships is mainly considered from linking relationship, the mobile phone communication on social network sites
And this kind of acquaintance's social relationships such as friend, the acquaintance of collected user such as short message record, cell phone address book record.And it is true
On, on the one hand, the development of social networks makes the more and more information of user be exposed in social networks, and sensitive information is exposed to
A variety of leakage of private information problems resulted in open social networks gradually cause people for the pass of secret protection subject under discussion
Note, this makes researchers be faced with more and more difficult and challenge in terms of obtaining user's social relationships data;Separately
On the one hand, in addition to these we can the close social relationships arrived of direct feel, there is also one in our daily life
The special social relationships of class, i.e. " known stranger ".Known stranger is such group, they can repeatedly phase
It meets, but they are not known each other each other from not to be noted other side, such as on the bus gone to work daily, in the body-building gone weekly yet
In shop, it is likely to encounter many strangers known in this way, it occupies the very big by one of the daily people that can be touched of people
Part, thus cannot be ignored.Liang in 2016 et al. exists《Europhysics Letters》On deliver
《Identifying Familiar Strangers in Human Encounter Networks》Known to one text devises
Stranger's grader is to strange relationship known to being excavated from different social relationships.It so far, there is not yet will
Such social relationships is applied to the correlative study work in position prediction field.
Invention content
Present invention aims at providing a kind of the location of mobile users prediction technique and device of automatic deduction social relationships,
To solve the social relationships data for needing to acquire user in existing position predicting method and consider user's social relationships insufficient
Problem improves the accuracy of position predicting method to protect the individual privacy of user.
The location of mobile users prediction technique of automatic deduction social relationships provided by the invention, the specific steps are:
(1) user's individual behavior record is obtained, i.e., from user's mobile behavior log database, obtains the individual row of user
For record, per data, record includes:User ID, access initial time, access duration, access place ID;
(2) social relationships type between deduction user infers the society between two users using user's individual behavior record
Can relationship type, social relationships type includes known stranger FS (familiar stranger), acquaintance F&IR (including friends
Friendly friend and colleague in-role), stranger S (stranger);
(3) user's social relation network is established, i.e., using user as node, the social relationships type between two users is to connect side
(not considering strange relationship) builds user's social relation network;
(4) user's motion track sequence is established, i.e., is recorded using user's individual behavior, structure user is in chronological order
Discrete motion track sequence;
(5) user group's social relationships die body is detected, i.e., generates social relationships subgraph using the German number of outstanding card;It is protected by spending
The method for staying cut edge reconnection generates randomized user social relation network, builds zero model;Compare each social relationships subgraph true
The magnitude relationship of statistical indicator z values under real network and zero model, determines user group's social relationships die body;
(6) user's individual social relationships die body is verified, and social relationships of the user is generated first with the German number of outstanding card
Figure;Compare whether user's individual social relationships subgraph is aforementioned social relationships die body, if then by verification, it is on the contrary then obstructed
It crosses;
(7) position forecaster is established, i.e., it is pre- to establish Markov forecast techniques device, acquaintance's fallout predictor, known stranger respectively
Survey device and output controller;If the individual is verified by social relationships die body, Markov forecast techniques device, known need to be only utilized
Stranger's fallout predictor and output controller predict the Future Positions of the user, if not verified, need markov pre-
Survey the Future Positions that device, acquaintance's fallout predictor, known stranger's fallout predictor and output controller predict the user.
In step (1) of the present invention, in the mobile behavior log database from user, the individual behavior note of user is obtained
Record, every record include:User ID, time, place, residence time.
In step (2) of the present invention, the social relationships type recorded using user's individual behavior between inferring two users, in detail
See Chinese patent " a kind of the line based on user's mobile behavior under social relationships sorting technique and device " (patent No.
201611264316.7), including:
It is recorded according to user behavior, obtains user and collect U, ground point set L.Per data, record includes User ID, access starting
Time, access duration, access place ID;
Time data in being recorded according to user behavior determines that user behavior cycle T, time discretization walk length Δ T,
In, the entire time shaft in daily record data is divided into N number of period by the user behavior cycle T;
For each period n, the behavioural matrix of structure user uWherein, u be user collect in U the
U user, n are n-th of period in N number of period,L indicates first of place in ground point set L;Behavioural matrix Sn
(u) element inIt is 0 or 1.
Indicate that user u and user the v l that is in the same localities possess the behavior record that the time overlaps with time and space co-occurrence.When
Empty co-occurrence represents user u and primary " alternative events " of the user v in real life.Define EnIt is all in n-th of period
The set of alternative events, if user u and user v n-th of period, place l, time step t once time and space co-occurrence, then
Alternative events en=(u, v, t, l) ∈ En。
The user of alternative events at least once is possessed to (u, v), structure Interactive matrix for every a pair
Wherein, u is u-th of user in user's collection U, and v is v-th of user in user's collection U,L is indicated in ground point set L
First of place.Interactive matrix Mu,vElementFor two tuples Indicate interaction weight,Indicate Cross support degree, whereinWithIt can be calculated by such as following formula (1), (2):
The rule degree d of user's space-time Interactive matrix is calculated by such as following formula (3)r(u,v):
The Spatial Temporal Entropy d of user's space-time Interactive matrix is calculated by such as following formula (4)e(u,v):
Build null hypothesis:User's individual behavior is not influenced by other people, and user's individual behavior does not have period skewed popularity.Root
According to null hypothesis, zero model of space-time Interactive matrix between user's individual behavior and user, i.e., the random user in each period are established
Behavioural matrix and random space-time Interactive matrix.
Individual liveness is calculated according to the user behavior matrix.User activity indicates that user accesses in one cycle
The probability of one space-time grid.User-space-time grid bigraph (bipartite graph) is established according to user behavior matrix;The user-space-time grid
Two part figures include:The user concentrates the node for indicating each user, indicates the node of each space-time grid (t, l) and deposits
Company side between the user and space-time grid of behavior record.Element in user behavior matrixWhen, user u and when
There is even side in empty grid (t, l).
Using company's edge flip method randomized user-space-time grid bigraph (bipartite graph) of reservation degree, random user-space-time grid is obtained
Bigraph (bipartite graph).The degree that this method retains each node is constant, and the quantity of node and Lian Bian are constant.
According to the individual liveness and the random user-space-time grid bigraph (bipartite graph), rebuild described in each period
Zero model of space-time Interactive matrix between user's individual behavior matrix and user, including:Random user behavioural matrixWith
Machine space-time Interactive matrixRandom law degreeWith random Spatial Temporal Entropy
The probability distribution of Spatial Temporal Entropy and rule degree in zero model is counted, and passes through preset Probability p0Determine Spatial Temporal Entropy and rule
Zero threshold value of degree, including:
Preset Probability p0, wherein p0Much smaller than 1.
According to the probability distribution of Spatial Temporal Entropy and rule degree in zero model, zero threshold value e of Spatial Temporal Entropy is determined0With rule degree zero
Threshold value r0.The wherein described zero threshold value e of Spatial Temporal Entropy0MeetZero threshold value r of the rule degree0Meet
By comparing real user Interactive matrix in two dimensions of Spatial Temporal Entropy and rule degree between its zero threshold value
Magnitude relationship determines social relationships under the line between two users (known stranger FS (familiar stranger), acquaintance F&
IR (including friend friend and colleague in-role), stranger S (stranger)), including:
If the Spatial Temporal Entropy of user's Interactive matrix is less than the random threshold value of Spatial Temporal Entropy, rule degree is more than the random threshold value of rule degree, then
Determine that social relationships are known stranger under line between user.If the Spatial Temporal Entropy of user's Interactive matrix is less than the random threshold of Spatial Temporal Entropy
Value, rule degree are more than the random threshold value of rule degree, it is determined that social relationships are known stranger FS under line between user.If user hands over
The Spatial Temporal Entropy of mutual matrix is more than the random threshold value of Spatial Temporal Entropy, it is determined that social relationships are acquaintance F&IR under line between user, wherein
If rule degree is more than the random threshold value of rule degree, it is determined that social relationships are colleague/classmate etc. in acquaintance under line between user
Occupational relation IR, if rule degree is less than the random threshold value of rule degree, it is determined that social relationships are in acquaintance under line between user
Friends F.
Described using user as node in step (3) of the present invention, the social relationships type between two users is that even side (does not consider
Strange relationship), user's social relation network is built, including:
Each user u is as node, and the social relationships type between two users is as even side e, if the society between two users
Relationship type is acquaintance or known strange relationship, then it is assumed that there are a company sides between the two users, connect side type and correspond to society
Thus the type of meeting relationship builds user's social relation network G=(U, ε), wherein U is that user gathers, and ε is to connect line set.
It is described to be recorded using user's individual behavior in step (4) of the present invention, build the chronological discrete movement of user
Track sets, including:
According to the time data in user behavior, time discretization step length Δ T is determined;For the behavior record of user,
If user there are a plurality of record, chooses access duration time longest or access times is most in time discretization step
Thus the place that place is walked as the time discretization builds the discrete motion track sequence of user.
It is described to generate social relationships subgraph using the German number of outstanding card in step (5) of the present invention, including:
The acquaintance neighborhood that Γ (u) and Γ (v) is respectively user u and v is defined, is calculated by such as following formula (5) every
A user and the outstanding person of its all social relationships block German number (Jarccard ' s coefficient) J:
The n rank social relationships subgraphs for defining user are preceding n most important (i.e. motion track is most like) for being predicted user
Social relationships individual type.
The social relationships of each user are sorted from big to small by the German number of outstanding person's card takes preceding n individual thus to obtain each use
The n rank social relationships subgraphs at family.
In step (5) of the present invention, the method that cut edge reconnection is retained by degree generates randomized user social relationships net
Network builds zero model, including:
A practical social relation network G=(U, ε) is given, social relationships are connected with a company side e in line set εst,
The company side of an identical social relationships type is randomly choosed, such as even sideThen randomly choose another
Lian Bian
With probabilityTwo company sides (u, v, FS) and (u ', v ', FS) are replaced with into (u, v ', FS) and (u ', v, FS), otherwise
They are replaced with into (u, u ', FS) and (v, v ', FS).If the process of the cut edge reconnection produce from ring while or weight while, terminate
The secondary cut edge reconnection operation.Above procedure is repeated until all even sides then obtain a random social relationships net by reconnection
Network.
100 random social relation networks as above are generated as zero model.
In step (5) of the present invention, more each social relationships subgraph statistical indicator z values under live network and zero model
Magnitude relationship, determine user group's social relationships die body, including:
Frequency of occurrences of the subgraph m in live network is indicated to Mr. Yu's subgraph type m, C (m),It is subgraph m true
Frequency of occurrence in random network corresponding to real network, μ (*) and σ (*) are respectively to calculate mean value and standard deviation operation, and definition is retouched
The index z values for stating subgraph importance are:
Z of each type of social relationships subgraph under the random network that live network and zero model generate is counted respectively
Value, if the z values of certain drawing of seeds are noticeably greater than 0, is confirmed as user group's social relationships die body.
In step (6) of the present invention, the social relationships subgraph that the user is generated using the German number of outstanding card, same to step (3),
Will the social relationships of each user by outstanding person block German number and sort from big to small preceding n individual is taken to obtain the n ranks society of each user
It can relationship subgraph.
In step (6) of the present invention, whether comparison user's individual social relationships subgraph is aforementioned social relationships die body,
If it is on the contrary then do not pass through then by verification, including:
The social relationships subgraph that the user generates is compared with the social relationships die body obtained in step (5), if the society
Can relationship subgraph be social relationships die body in step (5), then it is on the contrary then do not pass through verification by verification.
It is described to establish Markov forecast techniques device, acquaintance's fallout predictor, known stranger's fallout predictor in step (7) of the present invention
And output controller, including:
In Markov forecast techniques device, with stochastic variable XtIndicate place of the individual where discrete time walks t, with
All possible state { the x of machine variable1,x2,…,xt+1Can be detected from real data, each state xt∈{1,2,…,L}
For location number, L is the total number of different location, then the motion track of user is modeled with first order Markov chain,
That is next place of user only depends on the place of previous access, is represented by:
PM(Xt+1=xt+1|Xt=xt,Xt-1=xt-1,…,X1=x1)=PM(Xt+1=xt+1|Xt=xt) (7)
The single order Ma Erke that given user u walks the place of t-1 and extracted from history locality data in discrete time
Husband's transfer matrix can obtain the markov place access probability vector that the user u walks t in discrete time:
In acquaintance's fallout predictor, user is driven in a short time by its acquaintance F&IR towards its acquaintance institute
Position moved.
Assuming that the position that user u walks t+1 in discrete time will be predicted, some acquaintance v of user u is given discrete
The place l and v of time step t is always positioned at place l from time step t to t+1, usesIndicate that user u walks t+1 in discrete time
Access locations l,Indicate the probability that user u and user v meets in discrete time step t+1 in place l,Indicate that user v is always positioned at the probability of place l from time step t to t+1, then user u will in discrete time step t+1
The conditional probability of meeting access locations l is represented by:
The acquaintance set S of given user uF&IR={ v1,v2,…,vK, use wiIndicate user u and user viNormalizing
Change frequency of interactionProbability of so user u in discrete time step t+1 access locations l is represented by:
Obtain the probability that user u accesses each place in time step tAfterwards, user u can be obtained and walk t in discrete time
Place access probability vectorWherein normalization process will
By application to ensure
In known stranger's fallout predictor, due to the harmonic compoment that user interacts with its known stranger, user
Access locations can be showed by strange relationship reflex known to it.
As described in step (2), in each period n, the behavioural matrix of user u can be configured such thatIts
In, u is u-th of user in user's collection U, and n is n-th of period in N number of period,L is indicated in ground point set L
First of place.Behavioural matrix Sn(u) element inIt is 0 or 1.The accumulation behavioural matrix of user u is represented byWithIndicate user v time step t access locations l cumulative frequency, that
Probability of the user v in time step t access locations l can be expressed as:
Known stranger's set of relationship S of given user uFS={ v1,v2,…,vK, use wiIndicate user u and user vi
Normalization frequency of interactionProbability of so user u in discrete time step t access locations l is represented by:
Obtain the probability that user u accesses each place in time step tAfterwards, user u can be obtained and walk t's in discrete time
Place access probability vectorWherein normalization process will be by application with true
It protects
In output controller, to Markov forecast techniques device, acquaintance's fallout predictor and it is familiar with using multiple linear regression model
The output of stranger's fallout predictor be weighted fusion.If α, β and γ are weight parameter, and alpha+beta+γ=1, then final output
Place access probability vector PaggrIt is represented by:
Paggr=α PM+βPF&IR+γPFS (13)
Use PrealIndicate the actual place access probability vector of user, then weight parameter can be by minimizing loss function J
It obtains:
It, only need to can using Ma Er if it can be verified by the social relationships die body in inventive step (4) for user u
Husband's fallout predictor, known stranger's fallout predictor and output controller obtain finally predicting output, even parameter beta=0, otherwise need
Completely finally predicted using Markov forecast techniques device, acquaintance's fallout predictor, known stranger's fallout predictor and output controller
Output, i.e. parameter beta not necessarily 0.
On the other hand, the present invention also provides the automatic location of mobile users prediction meanss for inferring social relationships, including:
(1) user's individual behavior records acquisition module, for from user's mobile behavior log database, obtaining user
Body behavior record obtains user and collects U, ground point set L.Every user behavior record includes User ID, the time started, the duration,
Place.
(2) social relationships type inference module between user is established for being recorded using user's individual behavior between user
Space-time Interactive matrix extracts Spatial Temporal Entropy and rule degree, establishes zero model, determine zero threshold value by preset Probability p, the society between user
It can relationship deduction.
(3) user's social relation network establishes module, for using user as node, the social relationships type between two users to be
Even side (not considering strange relationship), user's social relation network is built.
(4) user's motion track sequence establishes module, and for being recorded using user's individual behavior, structure user is on time
Between sequence discrete motion track sequence.
(5) user group's social relationships die body detection module, for generating social relationships subgraph using the German number of outstanding card;It is logical
The method for crossing random cut edge reconnection generates randomized user social relation network, builds zero model;Compare each social relationships subgraph
The magnitude relationship of statistical indicator z values under live network and zero model, determines user group's social relationships die body.
(6) user's individual social relationships die body authentication module, the society for being generated the user using the German number of outstanding card are closed
It is subgraph;Compare whether user's individual social relationships subgraph is aforementioned social relationships die body, if then by verification, it is on the contrary then
Do not pass through.
(7) position forecaster establishes module, for establishing Markov forecast techniques device, acquaintance's fallout predictor, known footpath between fields respectively
Stranger's fallout predictor and output controller;If the individual is verified by social relationships die body, only need to utilize Markov forecast techniques device,
Known stranger's fallout predictor and output controller predict the Future Positions of the user, if not verified, need Ma Er
Can husband's fallout predictor, acquaintance's fallout predictor, known stranger's fallout predictor and output controller predict the Future Positions of the user.
Above-mentioned seven modules, the operation of specific seven steps for executing prediction technique of the present invention.Wherein:
User's individual social relationships die body authentication module, including:
Social relationships subgraph generates submodule, the social relationships subgraph for generating the user using the German number of outstanding card;
Social relationships die body verifies submodule, for comparing whether user's individual social relationships subgraph is that aforementioned society closes
It is die body, if then by verification, it is on the contrary then do not pass through.
The position forecaster establishes module, including:
Markov forecast techniques device setting up submodule, for predicting user's Future Positions using user's history location information;
Acquaintance's fallout predictor setting up submodule, for predicting user's Future Positions using the acquaintance of user;
Known stranger's fallout predictor setting up submodule, not for known stranger's Relationship Prediction user using user
Come position;
Output controller setting up submodule, for establishing son to Markov forecast techniques device setting up submodule, acquaintance's fallout predictor
The output of module and known stranger's fallout predictor setting up submodule is merged, and obtains finally predicting output.
Technical solution provided by the invention will have the following advantages:
The location of mobile users prediction technique and device of automatic deduction social relationships provided by the present invention are led in traditional
It crosses on the basis of user's history trace information and acquaintance's social relationships prediction user location, for the first time by " known stranger " relationship
Position predicting method is introduced, the accuracy of position prediction is improved, new thinking is provided for position predicting method design.This hair
It is bright by defining and excavating user's social relationships die body, the heterogeneous different types of user of processing calculates the same of cost reducing
The estimated performance of Shi Tigao entirety.The present invention is based on the different types of societies directly inferred using behavioral data under user's line
Relationship specially need not additionally acquire user's social relationships data, reduce the primary data information (pdi) needed for prediction, protect well
The individual privacy for having protected user reduces the difficulty of data acquisition.Method provided by the present invention is suitable for geographical point of Wi-Fi etc.
The higher scene of resolution is different from conventional method and depends on the lower scene of the geographical resolution ratio such as base station, POI more, can accomplish pair
It is preferably predicted in fine granularity geographical location.
Description of the drawings
Fig. 1 is a kind of stream of the location of mobile users prediction technique of automatic deduction social relationships provided in an embodiment of the present invention
Journey block diagram.
Fig. 2 is a kind of user's mobile behavior daily record data sample figure provided in an embodiment of the present invention.
Fig. 3 is a kind of user's social relationships type decision schematic diagram provided in an embodiment of the present invention.
Fig. 4 is that a kind of social relation network provided in an embodiment of the present invention generates schematic diagram.
Fig. 5 is a kind of user's social relationships subgraph schematic diagram provided in an embodiment of the present invention.
Fig. 6 is a kind of group of the location of mobile users prediction meanss of automatic deduction social relationships provided in an embodiment of the present invention
At structural schematic diagram.
The composed structure schematic diagram of Fig. 7 social relationships type inference modules between user provided in an embodiment of the present invention.
Fig. 8 is the composed structure schematic diagram that user's space-time Interactive matrix provided in an embodiment of the present invention establishes module.
Fig. 9 is the composed structure schematic diagram that zero model foundation provided in an embodiment of the present invention and zero threshold value choose module.
Figure 10 is the composed structure schematic diagram of user group's social relationships die body detection module provided in an embodiment of the present invention.
Figure 11 is the composed structure schematic diagram of user's individual social relationships die body authentication module provided in an embodiment of the present invention.
Figure 12 is the composed structure schematic diagram that position forecaster provided in an embodiment of the present invention establishes module.
Specific implementation mode
To make the purpose, technical scheme and advantage of the application be more clearly understood, below in conjunction with attached drawing, with it is domestic certain
Colleges and universities' wireless network logs in user behaviors log data instance, and present invention embodiment is described in detail.
Fig. 1 is the process blocks schematic diagram of the automatic location of mobile users prediction technique for inferring social relationships of the present invention, such as
Shown in Fig. 1, including:
Step 100, from user's mobile behavior log database, obtain user individual behavior record, including User ID,
Turn-on time, duration, place ID.
User behaviors log data instance is logged in the wireless network of domestic certain university acquisition, the wireless network in campus logs in row
Acquisition is done by school information for daily record and is stored, the wireless network of all users using Wireless LAN in Campus in campus is had recorded
Login behavior, raw data format are as shown in Figure 2.When per data, record includes that User ID, access initial time, access are lasting
Between, access place ID.It is concentrated in this sets of data, all different hotspot (AP) constitute ground point set.Due to wireless heat
Point coverage area is smaller, and user is often connected with it automatically apart from nearest hotspot, therefore, when user moves from one place
When to another place, the hotspot of access can also automatically switch.Every wireless network login record features user and connects
Enter the when and where of wireless network, and a series of wireless network login records of a user then feature the movement of the user
Behavior.
In the present embodiment, step 100 obtains user's individual behavior record, obtains from user's mobile behavior log database
Collect U, ground point set L to user.Per data, record includes User ID, access initial time, access duration, access place
ID.Raw data format is as shown in Fig. 2, every records with four-tuple (u, ta, δ t, l) form indicate, wherein u indicate user
Collect the Customs Assigned Number in U, taTo access initial time, δ t are the access duration, and l is place (the wireless heat in ground point set L
Point) number.
Step 101 records the social relationships type between inferring two users using user's individual behavior, including known
Stranger, acquaintance and stranger, refer to Chinese patent " under a kind of line based on user's mobile behavior social relationships sorting technique and
Device " (patent No. 201611264316.7), specifically comprises the following steps:
(1) it is recorded using user's individual behavior, establishes space-time Interactive matrix between user.
Time data in being recorded first according to user behavior determines that user behavior cycle T, time discretization walk length Δ
T, wherein the entire time shaft in daily record data is divided into N number of period by the user behavior cycle T.It is returned by counting user
The probability distribution (probability distribution is the probability distribution on entire user's collection) for returning the time interval in identical place, finds general
Notable time interval outstanding in rate, you can be considered as user behavior cycle T.In general, the behavior period of the mankind is 1 day or 7 days.
In the embodiment, T=7 days.The entire time shaft of observational record is divided into N number of period by T.On the other hand, in order to fully excavate
User moves the time of mobile behavior, and spatial model needs, by continuous time shaft discretization, to determine discrete so as to subsequent analysis
The expression of user's mobile behavior can be simplified by changing time step length Δ T, be the time that length is Δ T by continuous time discrete
Section.Depending on the selection of Δ T need to be according to specific data, it usually needs Δ T can remove some noises in data, and can fill
Divide the variation for showing user behavior.In this embodiment, Δ is taken T=3 hours.
For each period n, the behavioural matrix of structure user uWherein, n is N number of week described in epimere
N-th interim of period;T belongs toIndicate that t-th of time step in n-th of period, wherein Δ T are the time described in epimere
Step-length degree, a cycle is divided by itA time step;L indicates first of place in ground point set L.The user behavior matrix
Sn(u) line number is(the time step quantity in a cycle), columns are the place total quantity in ground point set L | L |.Sn(u) in
ElementIt is 1 or 0, when user u is happened at place l there are a behavior record, when the time step t in n-th of period,Otherwise,It should be noted that be equivalent to will be in a cycle for the user behavior matrix in a cycle
Time and space be divided intoA space-time grid, each space-time grid can indicate by two tuples (t, l),Table
Show that user has accessed space-time grid (t, l) within the period.
Space-time Interactive matrix between two two users is established according to time and space co-occurrence.The time and space co-occurrence indicates user u
Possess the behavior record that the time overlaps with the user v l that are in the same localities.Time and space co-occurrence represents user u and user v in reality
Primary " alternative events " in life.Define EnFor the set of all alternative events in n-th of period, if user u with use
Family v is n-th of period, place l, time step t once time and space co-occurrence, then alternative events en=(u, v, t, l) ∈ En。
The user of alternative events at least once is possessed to (u, v), structure Interactive matrix for every a pair
Wherein u is u-th of user in user collection U, and v is v-th of user in user collection U, and t belongs toIndicate the
T-th of time step in n period, l indicate first of place in ground point set L.The Interactive matrix Mu,vLine number be(one
Time step quantity in a period), columns is the place total quantity in ground point set L | L |.Interactive matrix Mu,vIt is equivalent to one
Time and space in period are divided intoA space-time grid, each space-time grid can be indicated by (t, l).Mu,vElementFor two tuples For interaction weight, indicate that user u and v is handed in space-time grid (t, l)
The number of cycles of mutual event,For Cross support degree, indicate user u and v when m- place grid (t, l) interact
The probability of event.WhereinWithIt can calculate in the following way:
Interaction weightWhen embodying two users (u, v) generation alternative events, to the preference journey of space-time grid (t, l)
Degree, Cross support degreeU is worked as in expression, and when v is mutual indepedent, the probability of an alternative events is generated in space-time grid (t, l).
When the behavior of user's space-time grid (t, l) is periodically stronger, Cross support degreeIt is then bigger.
(2) for the space-time Interactive matrix of each pair of user, two interaction characteristics are extracted:Spatial Temporal Entropy and rule degree.When wherein
Empty entropy is used to weigh the social similitude between two users, and rule degree is used to weigh the period of alternative events generation between two users
Change degree.
The rule degree d of user's space-time Interactive matrix is calculated in the following wayr(u,v):
The Spatial Temporal Entropy d of user's space-time Interactive matrix is calculated in the following waye(u,v):
(3) zero model is established, zero threshold value is determined by preset Probability p.
In order to distinguish different social relationships by two interaction characteristics of Spatial Temporal Entropy and rule degree, need to establish null hypothesis use
Zero model of space-time Interactive matrix between family obtains the Spatial Temporal Entropy under zero model and the distribution of rule degree.
By the randomization to user's individual behavior, space-time Interactive matrix between user's individual behavior and user is established
Zero model:Random user behavioural matrix in each period and random space-time Interactive matrix, it is true according to zero model and preset probability
Determine the random threshold value of Spatial Temporal Entropy and the random threshold value of rule degree, specifically comprises the following steps:
(a) individual liveness is calculated according to the user behavior matrix, the user activity indicates user in a week
The probability of a space-time grid is accessed in phase.User-space-time grid bigraph (bipartite graph) G is established according to the user behavior matrixUS, described
User-two part figure of space-time grid includes:The user concentrates the node for indicating each user, indicate each space-time grid (t,
L) node and there are the company sides between the user of behavior record and space-time grid.Element in the user behavior matrixWhen, there is even side with space-time grid (t, l) in user u.
When calculating individual liveness, it is all ground point set that user u was accessed to define L (u), in conjunction in step 100
User behavior matrix, user activity act (u) can calculate by following formula:
When establishing user-space-time grid bigraph (bipartite graph), the user behavior matrix under each period is traversed
If there are elementsThen there is even side in user u with space-time grid (t, l).
(b) company's edge flip method randomized user-space-time grid bigraph (bipartite graph) G of reservation degree is utilizedUS, obtain random user-when
Empty grid bigraph (bipartite graph)The degree that this method retains each node is constant, and the quantity of node and Lian Bian are constant.
In randomized user-space-time grid bigraph (bipartite graph), company's edge flip method of reservation degree is used.This method randomly selects two
Two company sides (u, (t1, l1)) in portion's figure, (v, (t2, l2)) is interacted, and obtains new company side (u, (t2, l2)), (v,
(t1, l1)), Xin Lianbian is added in bigraph (bipartite graph), and delete two original company sides.When the enough excessively secondary company's edge flips of progress
Afterwards, randomisation process is completed.User-space-time grid bigraph (bipartite graph) after randomization possesses number of nodes identical with artwork, even
Number of edges amount and node degree, that is to say, that each user node connects the space-time grid node of quantity identical as artwork, each space-time
Grid node connects user node identical with artwork quantity.Such method ensure that originally active node is still active,
Originally it is more to be still accessed quantity for the space-time grid more than accessed quantity.Random user-space-time grid bigraph (bipartite graph) is usedIt indicates.
In this step, the user of each user-space-time grid bigraph (bipartite graph) randomisation process is independent, be ensure that random
The space-time grid that user connects after change is not influenced by its social relationships, meets first hypothesis in null hypothesis.
(c) individual liveness and random user-space-time grid bigraph (bipartite graph) described in (b) are every with reconstruction according to (a)
The randomized model of space-time Interactive matrix between user's individual behavior matrix in a period and user, including:Random user
Behavioural matrixRandom space-time Interactive matrixRandom law degreeWith random Spatial Temporal Entropy
Establishing random user behavioural matrixWhen, for each period n, if random user-
Space-time grid bigraph (bipartite graph)It is middle to exist even side (u, (t, l)), thenMiddle elementIt is set to 1 with probability act (u), it is no
It is then 0.The step made under each period, and user is identical to the connection probability of each attachable space-time grid, and the period is not present
Property space-time the case where being biased to, meet second in null hypothesis hypothesis.
Establishing random space-time Interactive matrixWhen, initially set up the random alternative events collection in each periodIt is right
In random user behavioural matrixWithIn element, ifThen random alternative events
Correspondingly, being defined according to the Interactive matrix in (1), random Interactive matrix can be obtained
Mu,vElementFor two tuplesIt calculates as follows:
WhereinFor random user behavioural matrixElement
According to the interaction characteristic computational methods in (1), random law degree can be calculate by the following formulaWith
Random Spatial Temporal Entropy
WhereinFor random Interactive matrixElement.
(d) preset Probability p0, wherein p0Much smaller than 1.According to rule degree under zero model and random Spatial Temporal Entropy probability point
Cloth determines zero threshold value e of Spatial Temporal Entropy0With zero threshold value r of rule degree0.Wherein e0Meetr0MeetUsual p0Value be less than 0.001 to ensure enough confidence levels, work as p0When sufficiently small, meaning
Taste in the case of completely random, and the rule degree or Spatial Temporal Entropy of user's Interactive matrix are hardly possible to be more than corresponding to them
Zero threshold value, if the case where being more than zero threshold value occurs in interaction characteristic between user, be due between them in reality scene
Certain nonrandom social relationships caused by.
(4) big between its random threshold value in two dimensions of Spatial Temporal Entropy and rule degree by comparing user's Interactive matrix
Small relationship, including known stranger FS (familiar stranger), acquaintance F&IR (friend friend and occupational relation in-
Role), stranger S (stranger).
In the present embodiment, two users' social relationships type decision schematic diagram is as shown in Figure 3.
If the Spatial Temporal Entropy d of user's Interactive matrixe(u, v) is less than zero threshold value e of Spatial Temporal Entropy0, rule degree dr(u, v) is less than rule
Spend zero threshold value r0, it is determined that social relationships are strange relationship between user;If the Spatial Temporal Entropy of user's Interactive matrix is less than Spatial Temporal Entropy
Zero threshold value, rule degree are more than zero threshold value of rule degree, it is determined that social relationships are known strange relationship between user;If user hands over
The Spatial Temporal Entropy of mutual matrix is more than zero threshold value of Spatial Temporal Entropy, and rule degree is less than zero threshold value of rule degree, it is determined that between user social relationships be
Friends;If the Spatial Temporal Entropy of user's Interactive matrix is more than zero threshold value of Spatial Temporal Entropy, rule degree is more than zero threshold value of rule degree, it is determined that
Social relationships are the occupational relations such as colleague/classmate between user.
Step 102, using user as node, social relationships type between two users be even side (not considering strange relationship),
Build user's social relation network.
User is collected into each user u in U as node, utilizes the society between two users accessed in step 101
Meeting relationship type is as even side e.Only occurred one since the user for a pair of relationship strange each other is for, between them
Secondary interaction does not have substantive help for the position prediction of both sides, therefore will not consider strange relationship S here.Such as
Social relationships type between fruit two users is acquaintance F&IR or known strange relationship FS, then thinks between the two users
There are a social relationships to connect side, and the type on this company side corresponds to the type of two users' social relationships, thus builds user society
Meeting relational network G=(U, ε), wherein U are that user gathers, and ε is to connect line set.The generation exemplary plot of social relation network is shown in Fig. 4.
User behaviors log data instance is logged in the wireless network of domestic certain university acquisition, the social relation network scale of generation is 10146
A node, 5182743 company sides.
Step 103 is recorded using user's individual behavior, builds the chronological discrete motion track sequence of user
Row.According to described in step 101, time discretization step length Δ T can determine using user behavior data.In this embodiment, it takes
Δ T=3 hours.Recorded according to the primitive behavior of user, if user in time discretization step there are a plurality of record, select
The place for taking access duration time longest or the most place of access times to be walked as the time discretization.Such as certain user u exists
Respectively in place l in certain time discretization step1、l2And l3It accesses 53 minutes, 28 minutes and 1 hour, then building the user's
When discrete motion track sequence, access locations of the user in time discretization step choose l3.Thus each use is obtained
The discrete motion track sequence at family.In the present embodiment, since the time span of data set is 84 days
Step 104 generates social relationships subgraph using the German number of outstanding person's card;It is generated by the method for random cut edge reconnection random
Change user's social relation network, builds zero model;Compare each social relationships subgraph statistical indicator z under live network and zero model
The magnitude relationship of value determines user group's social relationships die body.
In the present embodiment, specifically comprise the following steps:
(1) social relationships subgraph is generated using the German number of outstanding card.The n rank social relationships subgraphs for defining user first are pre-
The social relationships individual type of preceding n most important (i.e. motion track is most like) of user is surveyed, exemplary plot is shown in Fig. 5.Fig. 5 is shown
Four kinds of different forms of 3 rank social relationships subgraphs connect wherein the Centroid of each subgraph is the user of position to be predicted
3 nodes connect are the user of first 3 most important (i.e. motion track is most like) of the position user to be predicted, even side type pair
Should user couple social relationships type.It is that one kind portraying social proximity between user in community network research that outstanding person, which blocks German number,
Representative index, apparent positive correlation is presented with the track similarity of user couple, while calculating and outstanding blocking German number and compare
The track similarity for directly calculating user couple has lower computation complexity.Therefore, excavate user's using the German number of outstanding card
Preceding n most important social relationships individuals.Define the acquaintance F&IR neighbours collection that Γ (u) and Γ (v) is respectively user u and v
It closes, calculates each user in the following way and the outstanding person of its all social relationships blocks German number (Jarccard ' s
coefficient)J:
The social relationships of each user are sorted from big to small by the German number of outstanding person's card takes preceding n individual thus to obtain each use
The n rank social relationships subgraphs at family.
(2) method for retaining cut edge reconnection by degree generates randomized user social relation network, builds zero model.It is given
One practical social relation network G=(U, ε), the generation step of zero model are represented by:
(a) social relationships are connected with a company side e in line set εst, one identical social relationships type of random selection
Lian Bian, such as even sideThen randomly choose another company side
(b) with probabilityTwo company sides (u, v, FS) and (u ', v ', FS) are replaced with into (u, v ', FS) and (u ', v, FS), it is no
They are then replaced with into (u, u ', FS) and (v, v ', FS);
If (c) process of step (b) cut edge reconnection produce from ring while or weight while, terminate the secondary cut edge reconnection operation.
Above procedure is repeated until all even sides then obtain a random social relation network by reconnection;
(d) 100 random social relation networks as above are generated as zero model.
(3) magnitude relationship of more each social relationships subgraph statistical indicator z values under live network and zero model is determined and is used
Family mass society relationship die body.In general, in live network and random network statistics numbers present significant difference subgraph
It is considered as die body, and z values are to portray whether subgraph number difference in live network and random network shows in die body research
A kind of representative index write.Frequency of occurrences of the subgraph m in live network is indicated to Mr. Yu's subgraph type m, C (m),
For the frequency of occurrence in subgraph m random networks corresponding to live network, μ (*) and σ (*) are respectively to calculate mean value and standard deviation
Operation, the index z values that definition describes subgraph importance are:
Z of each type of social relationships subgraph under the random network that live network and zero model generate is counted respectively
Value, if the z values of certain drawing of seeds are noticeably greater than 0, is confirmed as user group's social relationships die body.
In the present embodiment, by taking 3 rank social relationships subgraphs as an example (exemplary plot is shown in Fig. 5), 4 kinds of different types counting
Social relationships subgraph z values it is as shown in the table:
3 rank subgraphs | ① | ② | ③ | ④ |
Z values | 0.93 | -13.51 | 7.28 | 39.15 |
Since the z values of 3 rank social relationships subgraphs 3. and 4. are noticeably greater than 0, in the present embodiment, 3 rank social relationships
3. and 4. subgraph is the social relationships die body of user group, and 3 rank social relationships subgraphs are 3. and 4. by known strange
Relationship dominates (in 3 company sides of 3 rank subgraphs known strange relationship occupy the majority or all).
Step 105, the social relationships subgraph that the user is generated using the German number of outstanding card;Compare user's individual social relationships
Whether subgraph is aforementioned social relationships die body, if then by verification, it is on the contrary then do not pass through.
First with described in step 104, i.e., the social relationships of the user is sorted from big to small by the German number of outstanding person's card and take preceding n
Individual obtains the n rank social relationships subgraphs of each user, and in the present embodiment, n takes 3, that is, the 3 rank societies for generating the user close
It is subgraph.
By identified user group's social relationships die body in the 3 rank social relationships subgraphs of the above-mentioned user and step 104
(in the present embodiment be 3 rank social relationships subgraphs 3. with 4.) compare, if the 3 rank social relationships subgraphs of the user be society close
It is one kind in die body, then then otherwise verifying and not passing through or not verification.
Step 106 establishes Markov forecast techniques device, acquaintance's fallout predictor, known stranger's fallout predictor and output tune respectively
Save device;If the individual is verified by social relationships die body, Markov forecast techniques device, known stranger's fallout predictor need to be only utilized
The Future Positions of the user are predicted with output controller, if not verified, need Markov forecast techniques device, acquaintance's prediction
Device, known stranger's fallout predictor and output controller predict the Future Positions of the user.Specifically comprise the following steps:
(1) Markov forecast techniques device is established.Markov forecast techniques device predicts future using the location information of user's history
Position.In Markov forecast techniques device, as soon as the motion track sequence of user can be modeled with first order Markov chain,
It is to say that next access locations of user only rely upon the place of previous access.With stochastic variable XtIndicate an individual in from
Dissipate the place where time step t, all possible state { x of stochastic variable1,x2,…,xt+1Can be detected from real data,
Each state xt∈ { 1,2 ..., L } is location number, and L is the total number of different location, is represented by:
PM(Xt+1=xt+1|Xt=xt,Xt-1=xt-1,…,X1=x1)=PM(Xt+1=xt+1|Xt=xt)
The single order Ma Erke that given user u walks the place of t-1 and extracted from history locality data in discrete time
Husband's transfer matrix can obtain the markov place access probability vector that the user u walks t in discrete time
(2) acquaintance's fallout predictor is established.Acquaintance's fallout predictor is based on user with current towards its acquaintance in a short time
The fact that the notable tendency of position movement, is designed.Such as if some friend of certain user has a meal in dining room place, then
The user is likely to be moved to the position that its friends is currently located to have a meal together with its friend.
Assuming that the position that user u walks t+1 in discrete time will be predicted, some acquaintance v of user u is given discrete
The place l and v of time step t is always positioned at place l from time step t to t+1, usesIndicate that user u walks t+1 in discrete time
Access locations l,Indicate the probability that user u and user v meets in discrete time step t+1 in place l,Indicate that user v is always positioned at the probability of place l from time step t to t+1, then user u walks t+1 in discrete time
Will the conditional probability of access locations l be represented by:
The acquaintance set S of given user uF&IR={ v1,v2,…,vK, use wiIndicate user u and user viNormalizing
Change frequency of interactionProbability of so user u in discrete time step t+1 access locations l is represented by:
I.e. the user is the weighting that each of which friend relation is exerted one's influence in the probability of discrete time step t+1 access locations l
With.
Obtain the probability that user u accesses each place in time step tAfterwards, user u can be obtained and walk t in discrete time
Place access probability vectorIn the present embodiment, due to
The sum of probability vector not necessarily 1 will apply a normalization process to ensure herein for the convenience calculated later
(3) stranger's fallout predictor known to establishing.User interacts (i.e. geography meets) tool with its known strange relationship
There are significant periodic characteristics, they can carry out continually periodically interaction in certain fixed places, therefore, pass through user's
Known strange relationship group, can be reversed the partial movement trace information for reappearing the user.
As described in inventive step (2), in each period n, the behavioural matrix of user u can be configured such that
Wherein, u is u-th of user in user's collection U, and n is n-th of period in N number of period,L is indicated in ground point set L
First of place.Behavioural matrix Sn(u) element inIt is 0 or 1.The accumulation behavioural matrix of user u is represented byWithIndicate user v time step t access locations l cumulative frequency, that
Probability of the user v in time step t access locations l can be expressed as:
Known stranger's set of relationship S of given user uFS={ v1,v2,…,vK, use wiIndicate user u and user vi
Normalization frequency of interactionProbability of so user u in discrete time step t access locations l is represented by:
Obtain the probability that user u accesses each place in time step tAfterwards, user u can be obtained and walk t's in discrete time
Place access probability vectorIn the present embodiment, due to the probability to
The sum of amount not necessarily 1 will apply a normalization process to ensure herein for the convenience calculated later
(4) output controller is established.According to described in step 106 (1) (2) (3) part can get Markov forecast techniques device,
Three place access probability vectors of acquaintance's fallout predictor and known stranger's fallout predictor, in order to obtain final unique place visit
It asks that probability vector exports, needs to merge three above-mentioned place access probability vectors.
Using multiple linear regression model to Markov forecast techniques device, acquaintance's fallout predictor and known stranger's fallout predictor
Output is weighted fusion.If α, β and γ are weight parameter, and alpha+beta+γ=1, then final output place access probability is vectorial
PaggrIt is represented by:
Paggr=α PM+βPF&IR+βPFS
Use PrealIndicate the actual place access probability vector of user, then weight parameter can be by minimizing loss function J
It obtains:
It, only need to can using Ma Er if it can be verified by the social relationships die body in inventive step (4) for user u
Husband's fallout predictor, known stranger's fallout predictor and output controller obtain finally predicting output, even parameter beta=0, otherwise need
Completely finally predicted using Markov forecast techniques device, acquaintance's fallout predictor, known stranger's fallout predictor and output controller
Output, i.e. parameter beta not necessarily 0.
In the present embodiment, it is training set initial data in chronological sequence sequentially to be divided to 50%, and 50% is test set, instruction
Practice collection and obtain weight parameter α, β and γ for training, test set is used to test the performance of prediction technique.In the present embodiment, side
The evaluation index of method is predictablity rate, the predictablity rate ζ of user uuIt indicates, i.e.,:
If the ith prediction of wherein user u is correct, that is to say, that user u has accessed predicted place really, that
ηi=1, otherwise ηi=0.The present embodiment is each user the prediction that time span is 1 week.
The present embodiment designs 3 Baseline Methods to compare evaluation, respectively:
(a) Baseline Methods 1:First order Markov chain prediction technique
(b) Baseline Methods 2:In conjunction with the prediction technique of first order Markov chain and acquaintance
(c) Baseline Methods 3:In conjunction with the prediction technique of first order Markov chain and known strange relationship
The automatic location of mobile users prediction technique for inferring social relationships that the present embodiment is provided and 3 Baseline Methods
Prediction result is compared as follows:
It can be seen that the location of mobile users prediction technique of automatic deduction social relationships provided by the present invention compares 3 kinds of bases
Line method predictablity rate highest, effect are best.
For ease of preferably implementing the said program of the embodiment of the present invention, the phase for implementing said program is also provided below
Close device.
It please refers to shown in Fig. 6, a kind of location of mobile users of automatic deduction social relationships provided in an embodiment of the present invention is pre-
Device 600 is surveyed, may include:User's individual behavior record acquisition module 601, social relationships type inference module 602 between user,
User's social relation network establishes module 603, user's motion track sequence establishes module 604, user group's social relationships die body
Detection module 605, user's individual social relationships die body authentication module 606 and position forecaster establish module 607.
User's individual behavior records acquisition module 601, for from user's mobile behavior log database, obtaining user
Body behavior record obtains user and collects U, ground point set L.Every user behavior record includes User ID, the time started, the duration,
Place;
Social relationships type inference module 602 between user is established for being recorded using user's individual behavior between user
Space-time Interactive matrix extracts Spatial Temporal Entropy and rule degree, establishes zero model, determine zero threshold value by preset Probability p, the society between user
Meeting relationship type is inferred;
User's social relation network establishes module 603, for using user as node, the social relationships type between two users to be
Even side (not considering strange relationship), user's social relation network is built;
User's motion track sequence establishes module 604, and for being recorded using user's individual behavior, structure user is on time
Between sequence discrete motion track sequence;
User group's social relationships die body detection module 605, for generating social relationships subgraph using the German number of outstanding card;It is logical
The method for crossing random cut edge reconnection generates randomized user social relation network, builds zero model;Compare each social relationships subgraph
The magnitude relationship of statistical indicator z values under live network and zero model, determines user group's social relationships die body;
User's individual social relationships die body authentication module 606, the society for being generated the user using the German number of outstanding card are closed
It is subgraph;Compare whether user's individual social relationships subgraph is aforementioned social relationships die body, if then by verification, it is on the contrary then
Do not pass through;
Position forecaster establishes module 607, for establishing Markov forecast techniques device, acquaintance's fallout predictor, known footpath between fields respectively
Stranger's fallout predictor and output controller;If the individual is verified by social relationships die body, only need to utilize Markov forecast techniques device,
Known stranger's fallout predictor and output controller predict the Future Positions of the user, if not verified, need Ma Er
Can husband's fallout predictor, acquaintance's fallout predictor, known stranger's fallout predictor and output controller predict the Future Positions of the user.
In an embodiment of the present invention, it please refers to shown in Fig. 7, social relationships type inference module 602 between the user, wraps
It includes:
The foundation of space-time Interactive matrix and interaction characteristic extracting sub-module 6021 between user, for utilizing user's individual row
For record, space-time Interactive matrix between user is established, extracts Spatial Temporal Entropy and rule degree;
Zero model foundation and zero threshold value of interaction characteristic choose submodule 6022 and pass through preset Probability p for establishing zero model
Determine zero threshold value;
Social relationships type decision submodule 6023 under line between user, for by comparing the true Interactive matrix space-time of user
Magnitude relationship between entropy and rule degree and its zero threshold value, determines social relationships under the line between two users;
In an embodiment of the present invention, it please refers to shown in Fig. 8, the foundation of space-time Interactive matrix and interaction characteristic between the user
Extracting sub-module 6021, including:
User's alternative events setting up submodule 60211, for determining all alternative events between user according to space-time co-occurrence,
Establish alternative events set;
Space-time Interactive matrix setting up submodule 60212, for being established to two users for possessing alternative events at least once
Space-time Interactive matrix between user, wherein each matrix element is two tuples, the weight and probability of common description interaction;
Interaction characteristic extracting sub-module 60213 is used to extract interaction characteristic according to the space-time Interactive matrix between user, including
Spatial Temporal Entropy and rule degree;
In an embodiment of the present invention, it please referring to shown in Fig. 9, zero model foundation and zero threshold value choose module 6022,
Including:
User's individual behavior is randomized submodule 60221, for carrying out randomization to user behavior, is used at random
Family behavioural matrix;
Random space-time Interactive matrix setting up submodule 60222, for according between random user behavioural matrix resume user with
Machine space-time Interactive matrix;
Zero threshold value extracting sub-module 60223 of interaction characteristic counts it for extracting the Spatial Temporal Entropy under zero model and rule degree
Probability distribution, and pass through preset Probability p0Determine zero threshold value of zero threshold value of Spatial Temporal Entropy and rule degree;
In an embodiment of the present invention, it please refers to Fig.1 shown in 0, user group's social relationships die body detection module
605, including:
User's social relationships subgraph generates submodule 6051, for generating social relationships subgraph using the German number of outstanding card;
Zero model foundation submodule 6052, for generating randomized user social relationships by the method for random cut edge reconnection
Network builds zero model;
User's social relationships die body determination sub-module 6053, for more each social relationships subgraph in live network and zero mould
The magnitude relationship of statistical indicator z values under type, determines user group's social relationships die body;
In an embodiment of the present invention, it please refers to Fig.1 shown in 1, user's individual social relationships die body authentication module
606, including:
Social relationships subgraph generates submodule 6061, social relationships for generating the user using the German number of outstanding card
Figure;
Whether social relationships die body verifies submodule 6062, be aforementioned society for comparing user's individual social relationships subgraph
Can relationship die body, if then by verification, it is on the contrary then do not pass through;
In an embodiment of the present invention, it please referring to Fig.1 shown in 2, the position forecaster establishes module 607, including:
Markov forecast techniques device setting up submodule 6071, for utilizing user's history location information prediction user future position
It sets;
Acquaintance's fallout predictor setting up submodule 6072, for predicting user's Future Positions using the acquaintance of user;
Known stranger's fallout predictor setting up submodule 6073, for being used using known stranger's Relationship Prediction of user
Family Future Positions;
Output controller setting up submodule 6074, for the defeated of submodule 6071, submodule 6072 and submodule 6073
Go out to be merged, obtains finally predicting output.
By the previous embodiment description of this invention it is found that first from user's mobile behavior log database, obtain
The individual behavior of user records, and infers the social relationships type between two users;Build user's social relation network;Build user from
Dissipate motion track sequence;Social relationships subgraph is generated using the German number of outstanding person's card, randomized user is generated by random cut edge reconnection
Social relation network builds zero model, and user group's social relationships die body is excavated by z values;User's individual more to be predicted society
Can relationship subgraph whether be aforementioned social relationships die body, if then by verification, it is on the contrary then do not pass through;Establish Markov forecast techniques
Device, acquaintance's fallout predictor, known stranger's fallout predictor and output controller;If the individual is verified by social relationships die body,
Only Ma Er need to be needed if not verified using Markov forecast techniques device, known stranger's fallout predictor and output controller
It can husband's fallout predictor, acquaintance's fallout predictor, known stranger's fallout predictor and output controller.Automatic deduction society provided by the present invention
The location of mobile users prediction technique and device of meeting relationship, pass through user's history trace information and acquaintance's social relationships traditional
On the basis of predicting user location, " known stranger " relationship is introduced into position predicting method for the first time, improves position prediction
Accuracy, for position predicting method design provide new thinking.The present invention is by defining and excavating user's social relationships mould
Body, the heterogeneous different types of user of processing, whole estimated performance is improved while reducing calculating cost.The present invention is based on
The different types of social relationships directly inferred using behavioral data under user's line specially need not additionally acquire user society
Relation data reduces the primary data information (pdi) needed for prediction, protects the individual privacy of user well, reduce data and obtain
The difficulty taken.Method provided by the present invention is suitable for the higher scene of the geographical resolution ratio such as Wi-Fi, and it is more to be different from conventional method
Dependent on the lower scene of the geographical resolution ratio such as base station, POI, can accomplish preferably to predict fine granularity geographical location.
One of ordinary skill in the art will appreciate that implement the method for the above embodiments be can be with
Relevant hardware is instructed to complete by program.Based on this understanding, technical scheme of the present invention is substantially right in other words
The part that the prior art contributes can be expressed in the form of software products, which is stored in readable
In the storage medium taken, such as the floppy disk of computer, USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only Memory),
Random access memory (RAM, Random Access Memory), magnetic disc or CD etc., including some instructions use so that
One computer installation (can be personal computer, server or network equipment etc.) executes each embodiment institute of the present invention
The method stated.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain
Lid is within protection scope of the present invention.Therefore, protection scope of the present invention should be based on the protection scope of the described claims.
Claims (14)
1. a kind of location of mobile users prediction technique of automatic deduction social relationships, which is characterized in that the specific steps are:
(1) user's individual behavior record is obtained, i.e., from user's mobile behavior log database, obtains the individual behavior note of user
Record, per data, record includes User ID, access initial time, access duration, access place ID;
(2) social relationships type between deduction user infers the society pass between two users using user's individual behavior record
Set type, including known stranger FS, acquaintance F&IR, stranger S, acquaintance F&IR include friend friend and colleague in-
role;
(3) establish user's social relation network, i.e., using user as node, in addition to strange relationship, the social relationships between two users
Type is to connect side, builds user's social relation network;
(4) establish user's motion track sequence, i.e., using user's individual behavior record, structure user it is chronological from
Dissipate motion track sequence;
(5) user group's social relationships die body is detected, i.e., generates social relationships subgraph using the German number of outstanding card;Broken by spending to retain
The method of side reconnection generates randomized user social relation network, builds zero model;Compare each social relationships subgraph in true net
The magnitude relationship of statistical indicator z values under network and zero model, determines user group's social relationships die body;
(6) user's individual social relationships die body is verified, and the social relationships subgraph of the user is generated first with the German number of outstanding card;Than
Whether it is aforementioned social relationships die body compared with user's individual social relationships subgraph, if then by verification, it is on the contrary then do not pass through;
(7) position forecaster is established, i.e., establishes Markov forecast techniques device, acquaintance's fallout predictor, known stranger's fallout predictor respectively
And output controller;If the individual is verified by social relationships die body, Markov forecast techniques device, known strange need to be only utilized
People's fallout predictor and output controller predict the Future Positions of the user, if not verified, need Markov forecast techniques device,
Acquaintance's fallout predictor, known stranger's fallout predictor and output controller predict the Future Positions of the user.
2. prediction technique according to claim 1, which is characterized in that described to be remembered using user's individual behavior in step (2)
Record infers that the social relationships type between two users, detailed process are:
Recorded according to user behavior, obtain user collect U, ground point set L, per data record includes User ID, access initial time,
Access duration, access place ID;
Time data in being recorded according to user behavior determines that user behavior cycle T, time discretization walk length Δ T, wherein
Entire time shaft in daily record data is divided into N number of period by the user behavior cycle T;
For each period n, the behavioural matrix of structure user uWherein, u is u-th in user's collection U
User, n are n-th of period in N number of period,L indicates first of place in ground point set L;Behavioural matrix Sn(u)
In elementIt is 0 or 1;
Indicate that user u and user the v l that is in the same localities possess the behavior record that the time overlaps with time and space co-occurrence;Space-time is total
Represent user u and primary " alternative events " of the user v in real life;Define EnFor all interactions in n-th of period
The set of event, if user u and user v, n-th of period, place l, time step t once time and space co-occurrence then interact
Event en=(u, v, t, l) ∈ En;
The user of alternative events at least once is possessed to (u, v), structure Interactive matrix for every a pairWherein,
U is u-th of user in user's collection U, and v is v-th of user in user's collection U,L indicates the l in ground point set L
A place;Interactive matrix Mu,vElementFor two tuples Indicate interaction weight,Table
Show Cross support degree, whereinWithIt is calculated by such as following formula (1), (2):
The rule degree d of user's space-time Interactive matrix is calculated by such as following formula (3)r(u,v):
The Spatial Temporal Entropy d of user's space-time Interactive matrix is calculated by such as following formula (4)e(u,v):
Build null hypothesis:User's individual behavior is not influenced by other people, and user's individual behavior does not have period skewed popularity;According to zero
It is assumed that establish zero model of space-time Interactive matrix between user's individual behavior and user, i.e., the random user behavior in each period
Matrix and random space-time Interactive matrix;
Individual liveness is calculated according to the user behavior matrix;User activity indicates that user accesses one in one cycle
The probability of space-time grid;User-space-time grid bigraph (bipartite graph) is established according to user behavior matrix;The user-space-time grid two
Component includes:The user concentrates the node for indicating each user, indicates the node of each space-time grid (t, l) and there is row
For the company side between the user and space-time grid of record;Element in user behavior matrixWhen, user u and space-time grid
There is even side in lattice (t, l);
Using company's edge flip method randomized user-space-time grid bigraph (bipartite graph) of reservation degree, random user-space-time grid two is obtained
Figure;Here the degree for retaining each node is constant, and the quantity of node and Lian Bian are constant;
According to the individual liveness and the random user-space-time grid bigraph (bipartite graph), the user in each period is rebuild
Zero model of space-time Interactive matrix between individual behavior matrix and user, including:Random user behavioural matrixWhen random
Empty Interactive matrixRandom law degreeWith random Spatial Temporal Entropy
The probability distribution of Spatial Temporal Entropy and rule degree in zero model is counted, and passes through preset Probability p0Determine Spatial Temporal Entropy and rule degree
Zero threshold value, including:
Preset Probability p0, wherein p0Much smaller than 1;
According to the probability distribution of Spatial Temporal Entropy and rule degree in zero model, zero threshold value e of Spatial Temporal Entropy is determined0With zero threshold value of rule degree
r0;The wherein described zero threshold value e of Spatial Temporal Entropy0MeetZero threshold value r of the rule degree0Meet
By comparing the size in two dimensions of Spatial Temporal Entropy and rule degree between its zero threshold value of real user Interactive matrix
Relationship determines social relationships under the line between two users:Known stranger FS, acquaintance F&IR, stranger S, including:
If the Spatial Temporal Entropy of user's Interactive matrix is less than the random threshold value of Spatial Temporal Entropy, rule degree is more than the random threshold value of rule degree, it is determined that
Social relationships are known stranger under line between user;If the Spatial Temporal Entropy of user's Interactive matrix is less than the random threshold value of Spatial Temporal Entropy, rule
Rule degree is more than the random threshold value of rule degree, it is determined that social relationships are known stranger FS under line between user;If user interacts square
The Spatial Temporal Entropy of battle array is more than the random threshold value of Spatial Temporal Entropy, it is determined that social relationships are acquaintance F&IR under line between user;Wherein, if rule
Rule degree is more than the random threshold value of rule degree, it is determined that social relationships are colleague or the classmate's occupation pass in acquaintance under line between user
It is IR, if rule degree is less than the random threshold value of rule degree, it is determined that social relationships are that the friend in acquaintance is closed under line between user
It is F.
3. prediction technique according to claim 2, which is characterized in that build user's social relationships net described in step (3)
Network, including:
Each user u is as node, and the social relationships type between two users is as even side e, if the social relationships between two users
Type is acquaintance or known strange relationship, then it is assumed that there are a company sides between the two users, and even type correspondence in side is social closes
Thus the type of system builds user's social relation network G=(U, ε), wherein U gathers for user, and ε is to connect line set.
4. prediction technique according to claim 3, which is characterized in that remembered using user's individual behavior described in step (4)
Record builds the chronological discrete motion track sequence of user, including:
According to the time data in user behavior, time discretization step length Δ T is determined;For the behavior record of user, if with
There are a plurality of records in time discretization step at family, then choose access duration time longest or the most place of access times
As the place of time discretization step, the discrete motion track sequence of user is thus built.
5. prediction technique according to claim 4, which is characterized in that generated using the German number of outstanding card described in step (5)
Social relationships subgraph, including:
The acquaintance neighborhood that Γ (u) and Γ (v) is respectively user u and v is defined, calculates each user in the following way
Block German several J with the outstanding person of its all social relationships:
The n rank social relationships subgraphs for defining user are preceding n most important social relationships individual types for being predicted user;
The social relationships of each user are sorted from big to small by the German number of outstanding person's card takes preceding n individual thus to obtain each user's
N rank social relationships subgraphs.
6. prediction technique according to claim 5, which is characterized in that retain cut edge reconnection by degree described in step (5)
Method generate randomized user social relation network, build zero model, including:
A practical social relation network G=(U, ε) is given, social relationships are connected with a company side e in line set εst, at random
Select the company side of an identical social relationships type;I.e. for even sideThen randomly choose another company side
With probabilityTwo company sides (u, v, FS) and (u ', v ', FS) are replaced with into (u, v ', FS) and (u ', v, FS), otherwise by them
Replace with (u, u ', FS) and (v, v ', FS);If the process of the cut edge reconnection produce from ring while or weight while, terminate this time break
Side reconnection operation;Above procedure is repeated until all even sides then obtain a random social relation network by reconnection;
100 random social relation networks as above are generated as zero model.
7. prediction technique according to claim 6, which is characterized in that more each social relationships subgraph described in step (5)
The magnitude relationship of statistical indicator z values under live network and zero model determines user group's social relationships die body, including:
Frequency of occurrences of the subgraph m in live network is indicated to Mr. Yu's subgraph type m, C (m),It is subgraph m in live network
Frequency of occurrence in corresponding random network, μ (*) and σ (*) are respectively to calculate mean value and standard deviation operation, definition description subgraph
The index z values of importance are:
Z value of each type of social relationships subgraph under the random network that live network and zero model generate is counted respectively,
If the z values of certain drawing of seeds are noticeably greater than 0, it is confirmed as user group's social relationships die body.
8. prediction technique according to claim 7, which is characterized in that generated using the German number of outstanding card described in step (6)
The social relationships of each user are blocked German number by outstanding person and sorted from big to small by the social relationships subgraph of the user, same to step (3)
N individual obtains the n rank social relationships subgraphs of each user before taking.
9. prediction technique according to claim 8, which is characterized in that compare user individual society described in step (6)
Whether relationship subgraph is aforementioned social relationships die body, if then by verification, it is on the contrary then do not pass through, including:
The social relationships subgraph that the user generates is compared with the social relationships die body obtained in step (5), if the society closes
Be subgraph be social relationships die body in step (5), then it is on the contrary then do not pass through verification by verification.
10. prediction technique according to claim 9, which is characterized in that step establishes Markov forecast techniques described in (7)
Device, acquaintance's fallout predictor, known stranger's fallout predictor and output controller, including:
In Markov forecast techniques device, with stochastic variable XtIndicate place of the individual where discrete time walks t, it is random to become
Measure all possible state { x1,x2,…,xt+1Can be detected from real data, each state xt∈ 1,2 ..., and L } it is ground
Point number, L are the total number of different location is used then the motion track of user is modeled with first order Markov chain
Next place at family only depends on the place of previous access, is expressed as:
PM(Xt+1=xt+1|Xt=xt,Xt-1=xt-1,…,X1=x1)=PM(Xt+1=xt+1|Xt=xt)
Given user u turns in the first order Markov that discrete time walks the place of t-1 and is extracted from history locality data
Matrix is moved, the markov place access probability vector that the user u walks t in discrete time is obtained:
In acquaintance's fallout predictor, user is driven in a short time by its acquaintance F&IR towards where its acquaintance
It is moved position;
Assuming that the position that user u walks t+1 in discrete time will be predicted, some acquaintance v of user u is given when discrete
The place l and v of spacer step t is always positioned at place l from time step t to t+1, usesIndicate that user u is visited in discrete time step t+1
Ask place l,Indicate the probability that user u and user v meets in discrete time step t+1 in place l,Indicate that user v is always positioned at the probability of place l from time step t to t+1, then user u walks t+1 in discrete time
Will the conditional probability of access locations l be expressed as:
The acquaintance set S of given user uF&IR={ v1,v2,…,vK, use wiIndicate user u and user viNormalization hand over
Crossing over frequency:Probability of so user u in discrete time step t+1 access locations l is expressed as:
Obtain the probability that user u accesses each place in time step tAfterwards, the ground that user u walks t in discrete time is obtained
Point access probability vectorIt is normalized, that is, is ensured
In known stranger's fallout predictor, due to the harmonic compoment that user interacts with its known stranger, the visit of user
Ask that place strange relationship reflex known to it shows;
In each period n, the behavioural matrix of user u is configured toWherein, u is the u in user's collection U
A user, n are n-th of period in N number of period,L indicates first of place in ground point set L;Behavioural matrix Sn
(u) element inIt is 0 or 1;The accumulation behavioural matrix of user u is expressed asWithIndicate user v time step t access locations l cumulative frequency, then user v is time step t access locations l's
Probability is to be expressed as:
Known stranger's set of relationship S of given user uFS={ v1,v2,…,vK, use wiIndicate user u and user viReturn
One changes frequency of interaction:Probability of so user u in discrete time step t access locations l is expressed as:
Obtain the probability that user u accesses each place in time step tAfterwards, the place that user u walks t in discrete time is obtained
Access probability vectorIt is normalized, it is ensured that
In output controller, using multiple linear regression model to Markov forecast techniques device, acquaintance's fallout predictor and known footpath between fields
The output of stranger's fallout predictor is weighted fusion;If α, β and γ are weight parameter, and alpha+beta+γ=1, then final output place
Access probability vector PaggrIt is expressed as:
Paggr=α PM+βPF&IR+γPFS
Use PrealIndicate the actual place access probability vector of user, then weight parameter is obtained by minimizing loss function J:
For user u, if it can be verified by the social relationships die body in inventive step (4), need to only it use markov pre-
It surveys device, known stranger's fallout predictor and output controller to obtain finally predicting output, even parameter beta=0, otherwise need complete
It obtains finally predicting output using Markov forecast techniques device, acquaintance's fallout predictor, known stranger's fallout predictor and output controller,
That is parameter beta not necessarily 0.
11. a kind of location of mobile users of the automatic deduction social relationships based on one of the claim 1-10 prediction techniques is pre-
Survey device, which is characterized in that including:
(1) user's individual behavior records acquisition module, for from user's mobile behavior log database, obtaining user's individual row
For record, obtains user and collect U, ground point set L;Every user behavior record includes User ID, time started, duration, place;
(2) social relationships type inference module between user establishes space-time between user for being recorded using user's individual behavior
Interactive matrix extracts Spatial Temporal Entropy and rule degree, establishes zero model, zero threshold value is determined by preset Probability p, and society closes between user
System is inferred;
(3) user's social relation network establishes module, is used for using user as node, in addition to strange relationship, the society between two users
Meeting relationship type is to connect side, builds user's social relation network;
(4) user's motion track sequence establishes module, and for being recorded using user's individual behavior, structure user is temporally suitable
The discrete motion track sequence of sequence;
(5) user group's social relationships die body detection module, for generating social relationships subgraph using the German number of outstanding card;By with
The method of machine cut edge reconnection generates randomized user social relation network, builds zero model;Compare each social relationships subgraph true
The magnitude relationship of statistical indicator z values under real network and zero model, determines user group's social relationships die body;
(6) user's individual social relationships die body authentication module, social relationships for generating the user using the German number of outstanding card
Figure;Compare whether user's individual social relationships subgraph is aforementioned social relationships die body, if then by verification, it is on the contrary then obstructed
It crosses;
(7) position forecaster establishes module, for establishing Markov forecast techniques device, acquaintance's fallout predictor, known stranger respectively
Fallout predictor and output controller;If the individual is verified by social relationships die body, it need to only utilize Markov forecast techniques device, be familiar with
Stranger's fallout predictor and output controller predict the Future Positions of the user, if not verified, need markov
Fallout predictor, acquaintance's fallout predictor, known stranger's fallout predictor and output controller predict the Future Positions of the user;
This 7 modules correspond to the operation content of 7 steps of prediction technique.
12. prediction meanss according to claim 11, which is characterized in that user group's social relationships die body detection module,
Including:
User's social relationships subgraph generates submodule, for generating social relationships subgraph using the German number of outstanding card;
Zero model foundation submodule, for generating randomized user social relation network, structure by the method for random cut edge reconnection
Build zero model;
User's social relationships die body determination sub-module is counted for more each social relationships subgraph under live network and zero model
The magnitude relationship of index z values determines user group's social relationships die body.
13. prediction meanss according to claim 11, which is characterized in that user's individual social relationships die body authentication module,
Including:
Social relationships subgraph generates submodule, the social relationships subgraph for generating the user using the German number of outstanding card;
Social relationships die body verifies submodule, for comparing whether user's individual social relationships subgraph is aforementioned social relationships mould
Body, if then by verification, it is on the contrary then do not pass through.
14. prediction meanss according to claim 11, which is characterized in that position forecaster establishes module, including:
Markov forecast techniques device setting up submodule predicts user's Future Positions according to user's history location information;
Acquaintance's fallout predictor setting up submodule predicts user's Future Positions using the acquaintance of user;
Known stranger's fallout predictor setting up submodule, according to known stranger's Relationship Prediction user's Future Positions of user;
Output controller setting up submodule, for Markov forecast techniques device setting up submodule, acquaintance's fallout predictor setting up submodule
Output with known stranger's fallout predictor setting up submodule is merged, and obtains finally predicting output.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810222692.2A CN108650614B (en) | 2018-03-19 | 2018-03-19 | Mobile user position prediction method and device for automatically deducing social relationship |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810222692.2A CN108650614B (en) | 2018-03-19 | 2018-03-19 | Mobile user position prediction method and device for automatically deducing social relationship |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108650614A true CN108650614A (en) | 2018-10-12 |
CN108650614B CN108650614B (en) | 2020-07-28 |
Family
ID=63744298
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810222692.2A Active CN108650614B (en) | 2018-03-19 | 2018-03-19 | Mobile user position prediction method and device for automatically deducing social relationship |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108650614B (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110262855A (en) * | 2019-05-28 | 2019-09-20 | 东华大学 | Member in car networking based on background information speculates attack prototype system |
CN110633341A (en) * | 2019-01-15 | 2019-12-31 | 天津完美引力科技有限公司 | Apparatus and method for content recommendation based on time and geographic information |
CN111104609A (en) * | 2018-10-26 | 2020-05-05 | 百度在线网络技术(北京)有限公司 | Interpersonal relationship prediction method, interpersonal relationship prediction device, computer program, and storage medium |
CN111125551A (en) * | 2019-11-12 | 2020-05-08 | 杭州电子科技大学 | User position prediction method based on Markov model with selective memory |
CN111125272A (en) * | 2018-10-31 | 2020-05-08 | 百度在线网络技术(北京)有限公司 | Regional feature acquisition method and device, computer equipment and medium |
CN111667099A (en) * | 2020-05-18 | 2020-09-15 | 东北大学 | Dynamic target uncertain motion trajectory prediction method based on time granularity improvement |
CN113641917A (en) * | 2020-05-11 | 2021-11-12 | 杭州海康威视数字技术股份有限公司 | Relationship acquisition method and device |
CN113657635A (en) * | 2020-05-12 | 2021-11-16 | ***通信集团湖南有限公司 | Method for predicting communication user loss and electronic equipment |
CN115622973A (en) * | 2022-09-29 | 2023-01-17 | 中国人民解放军战略支援部队信息工程大学 | Stranger social instant messaging application user positioning method and device |
CN117014224A (en) * | 2023-09-12 | 2023-11-07 | 联通(广东)产业互联网有限公司 | Network attack defense method and system based on Gaussian process regression |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120323558A1 (en) * | 2011-02-14 | 2012-12-20 | Decisive Analytics Corporation | Method and apparatus for creating a predicting model |
CN104680250A (en) * | 2015-02-11 | 2015-06-03 | 北京邮电大学 | Position predicting system |
CN106326345A (en) * | 2016-08-08 | 2017-01-11 | 浙江工业大学 | Friend relation mining method based on user behaviors in social network |
CN106372072A (en) * | 2015-07-20 | 2017-02-01 | 北京大学 | Location-based recognition method for user relations in mobile social network |
CN106570764A (en) * | 2016-11-09 | 2017-04-19 | 广州杰赛科技股份有限公司 | User relationship predicting method and device |
CN106600052A (en) * | 2016-12-12 | 2017-04-26 | 西安交通大学 | User attribute and social network detection system based on space-time locus |
CN106682212A (en) * | 2016-12-31 | 2017-05-17 | 复旦大学 | Social relations classification method based on user movement behavior and device |
-
2018
- 2018-03-19 CN CN201810222692.2A patent/CN108650614B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120323558A1 (en) * | 2011-02-14 | 2012-12-20 | Decisive Analytics Corporation | Method and apparatus for creating a predicting model |
CN104680250A (en) * | 2015-02-11 | 2015-06-03 | 北京邮电大学 | Position predicting system |
CN106372072A (en) * | 2015-07-20 | 2017-02-01 | 北京大学 | Location-based recognition method for user relations in mobile social network |
CN106326345A (en) * | 2016-08-08 | 2017-01-11 | 浙江工业大学 | Friend relation mining method based on user behaviors in social network |
CN106570764A (en) * | 2016-11-09 | 2017-04-19 | 广州杰赛科技股份有限公司 | User relationship predicting method and device |
CN106600052A (en) * | 2016-12-12 | 2017-04-26 | 西安交通大学 | User attribute and social network detection system based on space-time locus |
CN106682212A (en) * | 2016-12-31 | 2017-05-17 | 复旦大学 | Social relations classification method based on user movement behavior and device |
Non-Patent Citations (7)
Title |
---|
BAUMANN P, KLEIMINGER W, SANTINI S: ""The influence of temporal and spatial features on the performance of next-place prediction algorithms"", 《PROCEEDINGS OF THE 2013 ACM》 * |
GAMBS S, KILLIJIAN M O, DEL PRADO CORTEZ M N: ""Next place prediction using mobility markov chains"", 《PROCEEDINGS OF THE FIRST WORKSHOP ON MEASUREMENT, PRIVACY, AND MOBILITY. ACM》 * |
GAO H, TANG J, HU X: ""Modeling temporal effects of human mobile behavior on location-based social networks"", 《ROCEEDINGS OF THE 22ND ACM INTERNATIONAL CONFERENCE ON》 * |
NOULAS A, SCELLATO S, LATHIA N: ""Mining User Mobility Features for Next Place Prediction in Location-Based Services"", 《2012 IEEE 12TH INTERNATIONAL CONFERENCE ON DATA MINING 》 * |
XIAO X,ZHENG Y: ""Inferring social ties between users with human location history "", 《JOURNAL OF AMBIENT INTELLIGENCE AND HUMANIZED COMPUTING》 * |
卢扬: ""人类移动行为模式研究"", 《中国优秀硕士学位论文全文数据库-社会科学Ⅱ辑》 * |
梁迪; 崔靖; 李翔: ""线下交互的动态社交网络研究进展:挑战与展望"", 《计算机学报》 * |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111104609A (en) * | 2018-10-26 | 2020-05-05 | 百度在线网络技术(北京)有限公司 | Interpersonal relationship prediction method, interpersonal relationship prediction device, computer program, and storage medium |
CN111104609B (en) * | 2018-10-26 | 2023-10-10 | 百度在线网络技术(北京)有限公司 | Inter-person relationship prediction method, inter-person relationship prediction device, and storage medium |
CN111125272A (en) * | 2018-10-31 | 2020-05-08 | 百度在线网络技术(北京)有限公司 | Regional feature acquisition method and device, computer equipment and medium |
CN110633341B (en) * | 2019-01-15 | 2023-07-14 | 北京完美知识科技有限公司 | Apparatus and method for content recommendation based on time and geographic information |
CN110633341A (en) * | 2019-01-15 | 2019-12-31 | 天津完美引力科技有限公司 | Apparatus and method for content recommendation based on time and geographic information |
CN110262855A (en) * | 2019-05-28 | 2019-09-20 | 东华大学 | Member in car networking based on background information speculates attack prototype system |
CN111125551A (en) * | 2019-11-12 | 2020-05-08 | 杭州电子科技大学 | User position prediction method based on Markov model with selective memory |
CN113641917A (en) * | 2020-05-11 | 2021-11-12 | 杭州海康威视数字技术股份有限公司 | Relationship acquisition method and device |
CN113657635B (en) * | 2020-05-12 | 2023-10-27 | ***通信集团湖南有限公司 | Method for predicting loss of communication user and electronic equipment |
CN113657635A (en) * | 2020-05-12 | 2021-11-16 | ***通信集团湖南有限公司 | Method for predicting communication user loss and electronic equipment |
CN111667099B (en) * | 2020-05-18 | 2023-10-10 | 东北大学 | Dynamic target uncertain motion trail prediction method based on time granularity improvement |
CN111667099A (en) * | 2020-05-18 | 2020-09-15 | 东北大学 | Dynamic target uncertain motion trajectory prediction method based on time granularity improvement |
CN115622973A (en) * | 2022-09-29 | 2023-01-17 | 中国人民解放军战略支援部队信息工程大学 | Stranger social instant messaging application user positioning method and device |
CN115622973B (en) * | 2022-09-29 | 2024-05-10 | 中国人民解放军战略支援部队信息工程大学 | Stranger social instant messaging application user positioning method and device |
CN117014224A (en) * | 2023-09-12 | 2023-11-07 | 联通(广东)产业互联网有限公司 | Network attack defense method and system based on Gaussian process regression |
CN117014224B (en) * | 2023-09-12 | 2024-01-30 | 联通(广东)产业互联网有限公司 | Network attack defense method and system based on Gaussian process regression |
Also Published As
Publication number | Publication date |
---|---|
CN108650614B (en) | 2020-07-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108650614A (en) | A kind of the location of mobile users prediction technique and device of automatic deduction social relationships | |
Zhang et al. | Comparison of machine learning algorithms for predicting crime hotspots | |
Gambs et al. | De-anonymization attack on geolocated data | |
Ye et al. | On the semantic annotation of places in location-based social networks | |
Cohen et al. | Rapid case-based mapping of seasonal malaria transmission risk for strategic elimination planning in Swaziland | |
CN106682212B (en) | Social relation classification method and device based on user mobile behaviors | |
Agarwal et al. | Data mining techniques for predicting dengue outbreak in geospatial domain using weather parameters for New Delhi, India | |
Ouyang et al. | Debiasing crowdsourced quantitative characteristics in local businesses and services | |
Chen et al. | Large-scale mobile fitness app usage analysis for smart health | |
Coppola et al. | Learning temporal context for activity recognition | |
Niu et al. | A real-time data collection mechanism with trajectory privacy in mobile crowd-sensing | |
Huang et al. | Mining location-based social networks for criminal activity prediction | |
Solomon et al. | Analyzing movement predictability using human attributes and behavioral patterns | |
Chen et al. | Mining spatiotemporal diffusion network: A new framework of active surveillance planning | |
Wan et al. | Addressing location uncertainties in GPS‐based activity monitoring: A methodological framework | |
Liu et al. | Coevil: A coevolutionary model for crime inference based on fuzzy rough feature selection | |
Grekousis et al. | Analyzing high-risk emergency areas with GIS and neural networks: The case of Athens, Greece | |
Nishino et al. | A location predictor based on dependencies between multiple lifelog data | |
Yang et al. | Spatio-temporal prediction of social connections | |
Zhou et al. | Multivariate Hawkes processes for incomplete biased data | |
CN116110588B (en) | Medical time sequence prediction method based on dynamic adjacency matrix and space-time attention | |
Doan et al. | Attractiveness versus competition: towards an unified model for user visitation | |
Günther et al. | Poster: Privacy-Preserving Epidemiological Modeling on Mobile Graphs | |
Almuammar et al. | Pattern discovery from dynamic data streams using frequent pattern mining with multi-support thresholds | |
Wang et al. | Combining Theory and Data-Driven Approaches for Epidemic Forecasts |
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 |