CN114880586A - Confrontation-based social circle inference method through mobility context awareness - Google Patents

Confrontation-based social circle inference method through mobility context awareness Download PDF

Info

Publication number
CN114880586A
CN114880586A CN202210636801.1A CN202210636801A CN114880586A CN 114880586 A CN114880586 A CN 114880586A CN 202210636801 A CN202210636801 A CN 202210636801A CN 114880586 A CN114880586 A CN 114880586A
Authority
CN
China
Prior art keywords
social circle
user
inference
track
context
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.)
Pending
Application number
CN202210636801.1A
Other languages
Chinese (zh)
Inventor
刘楠
张凤荔
赵芸伟
王瑞锦
高强
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN202210636801.1A priority Critical patent/CN114880586A/en
Publication of CN114880586A publication Critical patent/CN114880586A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9536—Search customisation based on social or collaborative filtering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/04—Inference or reasoning models
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/01—Social networking

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Databases & Information Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Marketing (AREA)
  • Primary Health Care (AREA)
  • Strategic Management (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Artificial Intelligence (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a social circle reasoning method based on confrontation through mobility context awareness. The method is characterized in that a novel end-to-end framework ASCI-CAM (adaptive-based Social circuits conference Context-Aware Mobility) is provided for deducing the Social relationship of the user from the Mobility data of the user. The ASCI-CAM utilizes a context map to obtain semantic track embedding from user sign-in behaviors, solves the TSCI (traditional-based social circle reference) problem in a mode of counterstudy, can well explain human movement patterns, and further improves the performance and the interpretability of a social circle reasoning model compared with the traditional method. Extensive experimental evaluation on a truly public data set (Brightkite, Gowalla and Foursquare) demonstrated the superior performance and interpretability of the results achieved by our method.

Description

Social circle inference method based on confrontation through mobility context awareness
Technical Field
The invention relates to the field of deep learning and social networks, in particular to a social circle reasoning method based on antagonism for mobility context awareness.
Background
The rapid development of wireless technology and smart devices has stimulated a substantial emergence of location-based social networking (lbs n) applications (e.g., Twitter, Instagram, Foursquare, and Yelp). However, in the real lbs ns, it is difficult to obtain any explicit information about the user identity due to privacy issues. Thus, social circle inference (TSCI) based on anonymous tracks is generated to infer social relationships by learning the movement patterns of individual users. Due to its importance in many LBSN applications, it has received widespread attention in recent years. In 2017, Yang et al measured the similarity of user preferences by finding the places users frequently visit and predicted social relationships. Duan et al formulated the friend recommendation problem as a fine-grained trajectory matching prediction problem to study spatiotemporal preferences of social relationships. However, these approaches still suffer from problems of (1) lack of modeling of contextual characteristics of user check-in; (2) structural information in user motion patterns cannot be captured; (3) the potential mobility distribution is not considered. These challenges greatly impact the performance and interpretability of existing models that solve TSCI problems.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a social circle inference method based on confrontation through mobility context awareness, semantic track embedding is obtained from user sign-in behaviors by utilizing a context map, the TSCI problem is solved in a confrontation learning mode, a human movement mode can be well explained, and the performance of social circle inference is further improved.
In order to achieve the purpose of the invention, the invention is realized by adopting the following technical scheme:
through a social circle inference method based on antagonism through mobility context awareness, the method is optimized and promoted from three aspects of user privacy protection, model inference performance and model interpretability at the same time by combining context characteristics of user check-in, structural information in a user motion mode and distribution of potential mobility of the user. The method comprises the following specific steps:
(1) inputting a user track, and constructing a context map according to sign-in historical data of a user;
(2) embedding sign-in information, namely generating a semantic library for learning sign-in expression by using a random walk strategy through a sampling track of a context map, and training an embedded model on the basis of the semantic library;
(3) inputting track data, representing hidden states of tracks through an encoder of a fusion attention mechanism, and approximating the distribution of the hidden states through a regularization strategy of antagonism to generate a more robust track representation;
(4) and (3) deducing the social circle, namely inputting a social network label space, and optimizing a training model and testing by combining an inference module with the aim of minimizing the cost between a real label and a predicted label.
Drawings
FIG. 1 is a diagram of a model architecture of the present invention;
FIG. 2 is a schematic diagram of the algorithm of the present invention;
FIG. 3 is a comparative experimental plot of the present invention.
Detailed Description
The following description of the embodiments of the present invention is provided to facilitate the understanding of the present invention by those skilled in the art, but it should be understood that the present invention is not limited to the scope of the embodiments, and it will be apparent to those skilled in the art that various changes may be made without departing from the spirit and scope of the invention as defined and defined in the appended claims, and all matters produced by the invention using the inventive concept are protected.
The method consists of four algorithms, and the specific construction process is as follows:
(1) building context graphs
Step1 inputting trajectory data sequence of a group of users
Figure BDA0003680660500000031
Wherein the content of the first and second substances,
Figure BDA0003680660500000032
representing a historical track generated by an anonymous user,
Figure BDA0003680660500000033
indicating a particular check-in, and entering a social circle tag U corresponding to the group of users i ，
Figure BDA0003680660500000034
Figure BDA0003680660500000035
Is the social circle label space.
Step2 order
Figure BDA0003680660500000036
For a collection of check-in data of users, each check-in
Figure BDA0003680660500000037
Considering a node of graph G, if a user moves from one node to another and the distance between two arbitrary nodes does not exceed 1km, an edge e is constructed between the two nodes, and graph G is the constructed context graph.
(2) Sign-in information embedding
And Step3, sampling the graph G according to a group of tracks with fixed length by adopting a random walk strategy, and generating a semantic library for learning sign-in expression.
Step4 the geographic location of c is related to c due to the check-in l The time correlation is c t Generation of geographic position embedding v of c using Skip-Gram algorithm g (c l ) And obtaining the time stamp embedding information v of c by allocating the access time to one of 24 hours t (c t ) V is to be g (c l ) And v t (c t ) Splicing to obtain a check-in embedded representation v c (c)，
Figure BDA0003680660500000038
Step5, obtaining the embedded representation of all check-ins according to the specific method of Step2 by using the semantic library generated in Step1, namely
Figure BDA0003680660500000039
Then all the sign-in information V is checked in c (c) Embedded in context graph G.
(3) User mobility pattern embedding
Step6 given a trajectory
Figure BDA00036806605000000310
Let the parameter set of the LSTM cell be W and b, let
Figure BDA0003680660500000041
Respectively an input gate, a forgetting gate, an output gate and a storage unit, and obtaining the hidden state of each sign-in c by the following formula calculation
Figure BDA0003680660500000042
The specific formula is as follows:
Figure BDA0003680660500000043
Figure BDA0003680660500000044
Figure BDA0003680660500000045
step7 hidden state for track T by encoder with fused attention mechanism
Figure BDA0003680660500000046
And (4) performing representation. The specific encoding process is as follows:
Figure BDA0003680660500000047
Figure BDA0003680660500000048
equations (5) and (6) can be simplified as,
Figure BDA0003680660500000049
step8 using the reconstruction error
Figure BDA00036806605000000410
Forcing state
Figure BDA00036806605000000411
The potential information from the track T is retained. The specific calculation formula is as follows,
Figure BDA00036806605000000412
step9 construction of composition parameters
Figure BDA00036806605000000413
Parameterized generator
Figure BDA00036806605000000414
Discriminator f parameterized by parameter ω ω And set both as fully connected networks and select the popular Wasserstein GAN (WGAN) to train both major modules. Then, a Gaussian noise u to N (0,1) is randomly sampled and input to the generator to obtain
Figure BDA00036806605000000415
Figure BDA00036806605000000416
Step10 calculation of
Figure BDA00036806605000000417
And
Figure BDA00036806605000000418
Figure BDA00036806605000000419
Figure BDA00036806605000000420
step12, adjusting parameters when
Figure BDA0003680660500000051
Figure BDA0003680660500000052
And
Figure BDA0003680660500000053
when the three are simultaneously minimum, the three are optimal
Figure BDA0003680660500000054
(3) Social circle inference
Step13 let the social circle inference module be
Figure BDA0003680660500000055
The above-mentioned given trajectory is T,
Figure BDA0003680660500000056
the real social circle of which is labeled U,
Figure BDA0003680660500000057
and through an inference module
Figure BDA0003680660500000058
Obtaining corresponding social circle prediction labels thereof
Figure BDA0003680660500000059
Figure BDA00036806605000000510
Step14, let epsilon represent the cross entropy loss function, and adjust the parameter omega to make
Figure BDA00036806605000000511
At the minimum, the temperature of the mixture is controlled,
Figure BDA00036806605000000512
step15 from
Figure BDA00036806605000000513
And
Figure BDA00036806605000000514
sequentially extracting users T i And corresponding social circle tag U i And steps 4 through 14 are repeated, with the parameters theta, phi, omega,
Figure BDA00036806605000000515
omega, output the final inference model
Figure BDA00036806605000000516

Claims (3)

1. The social circle inference method based on antagonism through mobility context awareness is characterized in that:
(1) when the preference of a user to POIs is captured, a context awareness information graph based on user check-in data is constructed in a graph embedding mode, and a new visual angle is provided for solving the problem of mobile data sparsity;
(2) when a human movement mode is represented, a track encoder with a fusion attention mechanism is adopted to extract a highly structured movement mode from given track data, and a regularization strategy of a countermeasure type is used for learning track representation, so that backward collapse in a subsequent reasoning process caused by the fact that a coding space of the track representation is not regularized is avoided;
(3) the social circle inference adopts an inference method based on learning track representation, the whole learning track representation process is unsupervised antagonistic representation learning, and potential factors influencing human movement patterns can be fully mined on the premise of protecting user privacy, so that the social circle inference has the capability of generating more approximate real data distribution.
2. The method for social circle inference based on antagonism through mobility context awareness according to claim 1, wherein the method comprises the following specific steps:
(1) inputting a user track, and constructing a context map according to sign-in historical data of a user;
(2) embedding sign-in information, namely generating a semantic library for learning sign-in expression by using a random walk strategy through a sampling track of a context map, and training an embedded model on the basis of the semantic library;
(3) inputting track data, representing hidden states of tracks through an encoder of a fusion attention mechanism, and approximating the distribution of the hidden states through a regularization strategy of antagonism to generate a more robust track representation;
(4) and (4) social circle inference, namely inputting a social network label space, and optimizing a training model and testing by combining an inference module with the aim of minimizing the cost between a real label and a predicted label.
3. The method for social circle inference based on antagonism through mobility context awareness according to claim 2, wherein the specific algorithm of the method comprises:
(1) building context graphs
Step1: inputting a set of trajectory data sequences of a user
Figure FDA0003680660490000021
Wherein the content of the first and second substances,
Figure FDA0003680660490000022
Figure FDA0003680660490000023
representing a historical trail generated by an anonymous user,
Figure FDA0003680660490000024
indicating a particular check-in, and entering a social circle tag U corresponding to the group of users i ，U i ∈U，
Figure FDA0003680660490000025
Is a social circle label space;
step2: order to
Figure FDA0003680660490000026
For a collection of check-in data of users, each check-in
Figure FDA0003680660490000027
Regarding as a node of graph G, if a user moves from one node to another node and the distance between two arbitrary nodes does not exceed 1km, an edge e is constructed between the two nodes, and then graph G is a constructed context graph;
(2) sign-in information embedding
Step3: sampling the graph G according to a group of tracks with fixed length by adopting a random walk strategy to generate a semantic library for learning sign-in expression;
step4: since the geographic location of the check-in c is related to c l The time correlation is c t Generation of geographic position embedding v of c using Skip-Gram algorithm g (c l ) And obtaining the time stamp embedding information v of c by allocating the access time to one of 24 hours t (c t ) V is to be g (c l ) And v t (c t ) Splicing to obtain a check-in embedded representation v c (c)，v c (c)＝[v t (c t )，v g (c l )]，
Figure FDA0003680660490000028
Step5: obtaining the embedded representation of all check-ins from the semantic library generated in Step1 according to the specific method of Step2, namely
Figure FDA0003680660490000029
Then all the sign-in information V is checked in c (c) Embedded in context graph G;
(3) user mobility pattern embedding
Step6: given a trajectory
Figure FDA00036806604900000210
Let the parameter set of the LSTM cell be W and b, let
Figure FDA00036806604900000211
Respectively an input gate, a forgetting gate, an output gate and a storage unit, and obtaining the hidden state of each sign-in c by the following formula calculation
Figure FDA00036806604900000212
The specific formula is as follows:
Figure FDA0003680660490000031
Figure FDA0003680660490000032
Figure FDA0003680660490000033
step7: tong (Chinese character of 'tong')Encoder with over-fusion attention mechanism, hidden state of track T
Figure FDA0003680660490000034
And (4) performing representation. The specific encoding process is as follows:
Figure FDA0003680660490000035
Figure FDA0003680660490000036
step8: using reconstruction errors
Figure FDA0003680660490000037
Forcing state
Figure FDA0003680660490000038
The potential information from the track T is retained,
Figure FDA0003680660490000039
is composed of
Figure FDA00036806604900000310
Step9: constructing a composition parameter
Figure FDA00036806604900000327
Parameterized generator
Figure FDA00036806604900000311
Discriminator f parameterized by parameter ω ω And set both as fully connected networks and select the popular Wasserstein GAN (WGAN) to train both major modules. Then, a Gaussian noise u to N (0,1) is randomly sampled and input to the generator to obtain
Figure FDA00036806604900000312
Step 10: computing
Figure FDA00036806604900000313
And
Figure FDA00036806604900000328
Figure FDA00036806604900000314
Figure FDA00036806604900000315
step 11: adjusting the parameters when
Figure FDA00036806604900000316
And
Figure FDA00036806604900000317
when the three are simultaneously minimum, the three are optimal
Figure FDA00036806604900000318
(3) Social circle inference
Step12: let the social circle inference module be
Figure FDA00036806604900000319
The above-mentioned given trajectory is T,
Figure FDA00036806604900000320
the real social circle label is U, and U belongs to U, and the real social circle label passes through the reasoning module
Figure FDA00036806604900000321
Obtaining corresponding social circle prediction labels thereof
Figure FDA00036806604900000322
Step13: let epsilon denote the cross-entropy loss function,
Figure FDA00036806604900000323
by adjusting the parameter omega, the
Figure FDA00036806604900000324
Minimum;
step14: from
Figure FDA00036806604900000325
And
Figure FDA00036806604900000326
sequentially extracting users T i And corresponding social circle tag U i And steps 4 through 13 are repeated, with the parameters theta, phi, omega,
Figure FDA0003680660490000041
omega, output the final inference model
Figure FDA0003680660490000042
CN202210636801.1A 2022-06-07 2022-06-07 Confrontation-based social circle inference method through mobility context awareness Pending CN114880586A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210636801.1A CN114880586A (en) 2022-06-07 2022-06-07 Confrontation-based social circle inference method through mobility context awareness

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210636801.1A CN114880586A (en) 2022-06-07 2022-06-07 Confrontation-based social circle inference method through mobility context awareness

Publications (1)

Publication Number Publication Date
CN114880586A true CN114880586A (en) 2022-08-09

Family

ID=82679316

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210636801.1A Pending CN114880586A (en) 2022-06-07 2022-06-07 Confrontation-based social circle inference method through mobility context awareness

Country Status (1)

Country Link
CN (1) CN114880586A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116415293A (en) * 2023-02-23 2023-07-11 山东省人工智能研究院 User private attribute anonymization method based on generation of countermeasure network

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110147892A (en) * 2019-02-20 2019-08-20 电子科技大学 Mankind's Move Mode presumption model, training method and estimation method based on variation track context-aware
CN110442781A (en) * 2019-06-28 2019-11-12 武汉大学 It is a kind of based on generate confrontation network to grade ranked items recommended method
CN110826698A (en) * 2019-11-04 2020-02-21 电子科技大学 Method for embedding and representing crowd moving mode through context-dependent graph
CN111738447A (en) * 2020-06-22 2020-10-02 东华大学 Mobile social network user relationship inference method based on spatio-temporal relationship learning
CN112035745A (en) * 2020-09-01 2020-12-04 重庆大学 Recommendation algorithm based on counterstudy and bidirectional long-short term memory network
CN113378074A (en) * 2021-06-10 2021-09-10 电子科技大学 Social network user trajectory analysis method based on self-supervision learning
CN113723075A (en) * 2021-08-28 2021-11-30 重庆理工大学 Specific target emotion analysis method for enhancing and counterlearning of fused word shielding data
CN113849725A (en) * 2021-08-19 2021-12-28 齐鲁工业大学 Socialized recommendation method and system based on graph attention confrontation network

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110147892A (en) * 2019-02-20 2019-08-20 电子科技大学 Mankind's Move Mode presumption model, training method and estimation method based on variation track context-aware
CN110442781A (en) * 2019-06-28 2019-11-12 武汉大学 It is a kind of based on generate confrontation network to grade ranked items recommended method
CN110826698A (en) * 2019-11-04 2020-02-21 电子科技大学 Method for embedding and representing crowd moving mode through context-dependent graph
CN111738447A (en) * 2020-06-22 2020-10-02 东华大学 Mobile social network user relationship inference method based on spatio-temporal relationship learning
CN112035745A (en) * 2020-09-01 2020-12-04 重庆大学 Recommendation algorithm based on counterstudy and bidirectional long-short term memory network
CN113378074A (en) * 2021-06-10 2021-09-10 电子科技大学 Social network user trajectory analysis method based on self-supervision learning
CN113849725A (en) * 2021-08-19 2021-12-28 齐鲁工业大学 Socialized recommendation method and system based on graph attention confrontation network
CN113723075A (en) * 2021-08-28 2021-11-30 重庆理工大学 Specific target emotion analysis method for enhancing and counterlearning of fused word shielding data

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
QIANG GAO等: "Adversity-Based Social Circles Inference via Context-Aware Mobility" *
李其娜;李廷会;: "基于深度学习的情境感知推荐系统研究进展" *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116415293A (en) * 2023-02-23 2023-07-11 山东省人工智能研究院 User private attribute anonymization method based on generation of countermeasure network
CN116415293B (en) * 2023-02-23 2024-01-26 山东省人工智能研究院 User private attribute anonymization method based on generation of countermeasure network

Similar Documents

Publication Publication Date Title
Feng et al. Poi2vec: Geographical latent representation for predicting future visitors
CN111400620B (en) User trajectory position prediction method based on space-time embedded Self-orientation
CN109635989B (en) Social network link prediction method based on multi-source heterogeneous data fusion
CN104462190A (en) On-line position prediction method based on mass of space trajectory excavation
Elbes et al. An indoor localization approach based on deep learning for indoor location-based services
Mao et al. Jointly contrastive representation learning on road network and trajectory
CN113068131B (en) Method, device, equipment and storage medium for predicting user movement mode and track
Li et al. Defending malicious check-in using big data analysis of indoor positioning system: An access point selection approach
Yin et al. Automerge: A framework for map assembling and smoothing in city-scale environments
Long et al. Location prediction for individual vehicles via exploiting travel regularity and preference
CN114880586A (en) Confrontation-based social circle inference method through mobility context awareness
Habibi et al. Human trajectory prediction using similarity-based multi-model fusion
Lin et al. Where to go? Predicting next location in IoT environment
Reddy et al. An enhanced travel package recommendation system based on location dependent social data
Pang et al. Intercity simulation of human mobility at rare events via reinforcement learning
Liu et al. Multilevel federated learning-based intelligent traffic flow forecasting for transportation network management
CN115082896A (en) Pedestrian trajectory prediction method based on topological graph structure and depth self-attention network
Wang et al. Traffic flow prediction based on spatiotemporal potential energy fields
Chu et al. Simulating human mobility with a trajectory generation framework based on diffusion model
Lv et al. Private cell-ID trajectory prediction using multi-graph embedding and encoder-decoder network
Wang et al. Trajectory distribution aware graph convolutional network for trajectory prediction considering spatio-temporal interactions and scene information
Chen et al. Q‐EANet: Implicit social modeling for trajectory prediction via experience‐anchored queries
Chow et al. Representation learning of pedestrian trajectories using actor-critic sequence-to-sequence autoencoder
Luan et al. An efficient target tracking approach through mobile crowdsensing
He et al. GC-Loc: A graph attention based framework for collaborative indoor Localization using infrastructure-free signals

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
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20220809

WD01 Invention patent application deemed withdrawn after publication