CN110189167B - Mobile advertisement fraud detection method based on heterogeneous graph embedding - Google Patents

Mobile advertisement fraud detection method based on heterogeneous graph embedding Download PDF

Info

Publication number
CN110189167B
CN110189167B CN201910417284.7A CN201910417284A CN110189167B CN 110189167 B CN110189167 B CN 110189167B CN 201910417284 A CN201910417284 A CN 201910417284A CN 110189167 B CN110189167 B CN 110189167B
Authority
CN
China
Prior art keywords
mobile
fraud detection
node
application
model
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.)
Active
Application number
CN201910417284.7A
Other languages
Chinese (zh)
Other versions
CN110189167A (en
Inventor
胡金龙
庄懿
陈浪
黄旸珉
黄松
董守斌
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.)
South China University of Technology SCUT
Original Assignee
South China University of Technology SCUT
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 South China University of Technology SCUT filed Critical South China University of Technology SCUT
Priority to CN201910417284.7A priority Critical patent/CN110189167B/en
Publication of CN110189167A publication Critical patent/CN110189167A/en
Application granted granted Critical
Publication of CN110189167B publication Critical patent/CN110189167B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0242—Determining effectiveness of advertisements
    • 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
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0248—Avoiding fraud

Abstract

The invention discloses a mobile advertisement fraud detection method based on heterogeneous graph embedding, which comprises the following steps: 1) acquiring mobile advertisement log data and preprocessing the data; 2) extracting incidence relation data of a user, an application and an advertisement, and constructing a right heterogeneous graph; 3) defining a meta-path, setting the number of wandering times and the longest step length of each node, traversing nodes of the authorized heterogeneous graph, and constructing a node meta-path random wandering sequence; 4) constructing a low-dimensional dense vector representation of nodes in the weighted heterogeneous graph by using a language model; 5) defining a label to form tested data; 6) constructing a mobile advertisement fraud detection model; 7) inputting the tested data of the mobile application of the training part into a mobile advertisement fraud detection model for training to obtain a mobile advertisement fraud detection model; 8) and carrying out fraud detection on the mobile application by adopting a mobile advertisement fraud detection model. The invention effectively detects the fraudulent mobile application by utilizing the entity incidence relation in the mobile advertisement system.

Description

Mobile advertisement fraud detection method based on heterogeneous graph embedding
Technical Field
The invention relates to the technical field of mobile application advertisement fraud, in particular to a mobile advertisement fraud detection method based on heterogeneous graph embedding.
Background
As a novel marketing mode depending on an intelligent terminal, the mobile advertisement has the characteristics of accuracy, interactivity, flexibility, individuation and the like compared with the traditional media. However, the continuous growth of advertisement fraud poses a serious threat to the mobile advertisement market, it is very difficult to identify the fraud of mobile applications, and advertisement fraud detection has become a hot problem to be solved urgently in the mobile internet advertisement ecosystem. Graph analysis methods based on graph structure data are applied to anomaly and fraud detection due to good representation capability and robustness of the structured data.
The traditional analysis method based on the graph structure has low efficiency in large-scale graphs, the existing effective schemes such as deep learning are difficult to be directly applied to analysis of graph structure data, and the graph embedding method learns effective vector representation in a low-dimensional space for nodes in the graphs, so that subsequent graph data analysis is better supported. Aiming at a complicated and variable mobile advertisement fraud means, how to utilize a graph embedding-based method to carry out efficient detection on fraud mobile application is a problem to be solved urgently.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a mobile advertisement fraud detection method based on heterogeneous graph embedding, which can improve the accuracy of mobile application advertisement fraud detection.
In order to achieve the purpose, the technical scheme provided by the invention is as follows: a mobile advertisement fraud detection method based on heterogeneous graph embedding comprises the following steps:
1) acquiring mobile advertisement log data, and preprocessing the data;
2) extracting incidence relation data of users, applications and advertisements in the mobile advertisement ecosystem, and constructing an authorized heterogeneous graph and a meta template corresponding to the authorized heterogeneous graph
Figure BDA0002064836210000021
Figure BDA0002064836210000022
And
Figure BDA0002064836210000023
respectively represent a category set and a relation category set of edges and satisfy
Figure BDA0002064836210000024
3) Defining meta-paths
Figure BDA0002064836210000025
Setting the number of wandering times n and the longest step length l of each node, traversing the nodes in the weighted heterogeneous graph G, and constructing n weighted random wandering paths S of the nodes vv＝{Sv1,Sv2,...,SvnFourthly, finally obtaining an element path random walk sequence S with the right different composition G;
4) constructing a language model, and learning d-dimensional space dense vector representation X belonging to R in P mobile application nodes in the weighted abnormal graph GP×dTo form a fluid delivery systemInputting a feature vector;
5) manually marking the mobile application of the training part, and setting a label value of each mobile application according to the information of whether the mobile application is a fraud application; the tag of the fraudulent application is set to 1 and the tag of the non-fraudulent application is set to 0, resulting in PtrainIndividual label data
Figure BDA0002064836210000026
Ptrain＜P，PtrainApplying a total number P eta for the training part, wherein eta is a floating point number which is less than 1 but more than 0, and then combining the floating point number with the corresponding input feature vector in the step 4) to form tested data;
6) constructing a fraud detection model for detecting fraudulent mobile applications;
7) inputting the tested data into a fraud detection model, and acquiring parameters of the fraud detection model to obtain a mobile advertisement fraud detection model;
8) inputting the input characteristics of the mobile application which is not marked into the mobile advertisement fraud detection model to detect the fraud application.
In the step 1), the data preprocessing comprises data cleaning and missing value filling; the mobile ad log data contains four attributes: a. unique identification attribute: unique identifiers of users, applications, advertisements; b. the time attribute is as follows: the user uses the application to operate the specific time of the advertisement, and the time is accurate to the second; c. position attribute: identifying a geographic location at which the user is located; d. the device attribute is as follows: the model of the device used by the user, the size of the display screen, the operating system.
In step 2), the authorized heterogeneous graph G comprises three types of nodes which are user nodes respectively
Figure BDA0002064836210000031
Application node
Figure BDA0002064836210000032
And advertising node
Figure BDA0002064836210000033
The authorized different composition G comprises three node relations which are respectively used by users
Figure BDA0002064836210000034
User operated advertisement
Figure BDA0002064836210000035
Application display advertisement
Figure BDA0002064836210000036
Its corresponding meta template
Figure BDA0002064836210000037
Indicating that there is a mapping function for any node V ∈ V
Figure BDA0002064836210000038
And there is a mapping function for any connected edge
Figure BDA0002064836210000039
The weight of the edge between two adjacent nodes in the weighted abnormal graph G is determined by the corresponding operation information.
In step 3), a random walk sequence S of each node is constructedti1, 2., n, the sampling mode of the node is divided into two stages: the initial stage and the subsequent stage represent respectively the walk sequence StiLength of 0 to half element path
Figure BDA00020648362100000310
In the migration stage and length between
Figure BDA00020648362100000311
To the migration phase between the longest migration step l.
Further, the walk probability of the initial stage of constructing the random walk sequence of each node is:
Figure BDA00020648362100000312
wherein the content of the first and second substances,
Figure BDA00020648362100000313
and vi+1∈Vt+1Respectively a current node and a next node,
Figure BDA00020648362100000314
belongs to V for the current nodet+1A set of neighbor nodes of a type that,
Figure BDA00020648362100000315
for meta-paths, φ () is a node type mapping function.
Further, the walk probability of the subsequent stage of the random walk sequence of each node is:
Figure BDA00020648362100000316
wherein the content of the first and second substances,
Figure BDA00020648362100000317
and vi+1∈Vt+1Respectively a current node and a next node,
Figure BDA00020648362100000318
is the type of relationship with the current node and the next node, wiIs a relationship of
Figure BDA0002064836210000041
The weight of the last side of (b), beta is the offset,
Figure BDA0002064836210000042
for a set of neighbor nodes that are eligible,
Figure BDA0002064836210000043
in order to be a meta-path,
Figure BDA0002064836210000044
the function is mapped for edge type.
In the step 4), the constructed language model is a Skip-gram model and is accelerated by using a negative sampling mode, and the number of negative samples of negative sampling is fn；
The optimization function in the constructed Skip-gram model is as follows:
Figure BDA0002064836210000045
wherein v istFor given node v heterogeneous context
Figure BDA0002064836210000046
In (3), theta is a parameter of the model, Xv,
Figure BDA0002064836210000047
For nodes v and vtCorresponding low-dimensional node vector representation, XaIs a low-dimensional vector representation of any node in the graph.
Obtaining a dense vector representation in a low-dimensional space of nodes in the graph as
Figure BDA0002064836210000048
Where d is the vector dimension.
In step 6), the constructed fraud detection model is a classifier model, including a traditional machine learning model and a deep learning model.
Compared with the prior art, the invention has the following advantages and beneficial effects:
the invention captures richer structure and semantic relation by constructing a plurality of different composition graphs representing entity incidence relation in the mobile advertisement system; meanwhile, aiming at random walk of the meta-paths of a plurality of different composition graphs, the relationship between nodes with similar behaviors is tighter by adding weight constraint in the node propagation probability, so that the behavior information of the nodes can be better reflected by the vector obtained by embedding the graph, and the fraudulent mobile application can be effectively detected.
Drawings
FIG. 1 is a detailed flow chart of the method of the present invention.
FIG. 2 is a diagram of an all-rights heterogeneous graph.
Detailed Description
The present invention will be described in further detail with reference to examples and drawings, but the present invention is not limited thereto.
As shown in fig. 1, the method for detecting fraud in mobile advertisement based on heterogeneous graph embedding provided in this embodiment includes the specific steps of:
1) and acquiring mobile advertisement log data and preprocessing the data.
In this embodiment, the data preprocessing includes data cleaning and missing value filling; the mobile ad log data contains four attributes: a. unique identification attribute: a unique identifier of a user, application, advertisement, etc.; b. the time attribute is as follows: the user uses the application to operate the specific time of the advertisement, and the time is accurate to the second; c. position attribute: identifying the geographical location of the user, such as the country and city of the user, the IP address used by the user, and the like; d. the device attribute is as follows: the model of the device used by the user, the size of the display screen, the operating system, etc. For example, user a clicked on advertisement D on mobile application C using device B at a certain point in time.
2) Extracting the incidence relation data of users, applications and advertisements in the mobile advertisement ecosystem, and constructing a meta template corresponding to a weighted heterogeneous graph G ═ V, E, M-
Figure BDA0002064836210000051
The heterogeneous graph of all rights is shown in figure 2.
In this embodiment, the authorized configuration graph G includes three types of nodes, which are user nodes
Figure BDA0002064836210000052
Application node
Figure BDA0002064836210000053
And advertising node
Figure BDA0002064836210000054
Further, the weighted heterogeneous graph G includes three node relationships, which are respectively used by the user to apply
Figure BDA0002064836210000055
User operated advertisement
Figure BDA0002064836210000056
Application display advertisement
Figure BDA0002064836210000057
Its corresponding meta template
Figure BDA0002064836210000058
Indicating that there is a mapping function for any node V ∈ V
Figure BDA0002064836210000059
And there is a mapping function for any connected edge
Figure BDA00020648362100000510
Further, in the present invention,
Figure BDA00020648362100000511
and
Figure BDA00020648362100000512
respectively represent a category set and a relation category set of edges and satisfy
Figure BDA00020648362100000513
Furthermore, the weight of the edge between two neighboring nodes in the weighted differential graph is determined by the corresponding operation information, i.e. the weight of the edge between two neighboring nodes is determined by the ratio of the number of operation actions to the total number of operation actions.
3) The meta path is defined as follows:
Figure BDA0002064836210000061
setting the number of wandering times of each node as 30 and the longest step length as 40, traversing the nodes in the authorized heterogeneous composition, constructing 30 authorized random wandering paths of each node, and finally obtaining a meta-path random wandering sequence of the authorized heterogeneous composition;
in the present embodiment, a random walk sequence S of each node is constructedvi1, 2., the sampling mode of the node in 30 is divided into two stages: an initial stage and a subsequent stage representing a walk stage in which the length of the walk sequence is between 0 and length 2 of a half-element path and a walk stage in which the length is between 2 and the longest walk step 40, respectively;
further, the walk probability of the initial walk stage of the random walk sequence for each node is constructed as follows:
Figure BDA0002064836210000062
wherein the content of the first and second substances,
Figure BDA0002064836210000063
and vi+1∈Vt+1Respectively a current node and a next node,
Figure BDA0002064836210000064
belongs to V for the current nodet+1A set of neighbor nodes of a type that,
Figure BDA0002064836210000065
for meta-paths, φ () is a node type mapping function.
Further, the walk probability of the subsequent stage of the random walk sequence of each node is:
Figure BDA0002064836210000066
wherein the content of the first and second substances,
Figure BDA0002064836210000067
and vi+1∈Vt+1Respectively a current node and a next node,
Figure BDA0002064836210000068
is the type of relationship with the current node and the next node, wiIs a relationship of
Figure BDA0002064836210000069
The weight of the last side of (b), beta is the offset,
Figure BDA00020648362100000610
for a set of neighbor nodes that are eligible,
Figure BDA00020648362100000611
in order to be a meta-path,
Figure BDA00020648362100000612
the function is mapped for edge type.
4) And (4) constructing a language model, learning the low-dimensional dense vector representation of each mobile application node in the weighted abnormal graph, and constructing an input feature vector.
Furthermore, the constructed language model is a Skip-gram model and is accelerated by using a negative sampling mode, and the number of negative samples of the negative sampling is fnIn this embodiment, the number of negative samples of the negative sample is 5;
the optimization function in the Skip-gram model is as follows:
Figure BDA0002064836210000071
wherein v istFor given node v heterogeneous context
Figure BDA0002064836210000072
In (3), theta is a parameter of the model, Xv,
Figure BDA0002064836210000073
For nodes v and vtCorresponding low-dimensional node vector representation, XaIs a low-dimensional vector representation of any node in the graph.
Finally, the dense vector of the nodes in the graph in the low-dimensional space is expressed as
Figure BDA0002064836210000074
Where d is the vector dimension.
5) Manually marking the mobile application of the training part, and setting a label value of each mobile application according to the information of whether the mobile application is a fraud application; the tag of the fraudulent application is set to 1 and the tag of the non-fraudulent application is set to 0, resulting in PtrainIndividual label data
Figure BDA0002064836210000075
Ptrain＜P，PtrainApplying the total number P eta for the training part, and combining the total number P eta with the corresponding input feature vector in the step (4) to form tested data;
in this embodiment, η is 0.8.
6) A fraud detection model is constructed for detecting fraudulent mobile applications.
In this embodiment, the constructed fraud detection model is a random forest classifier model, and the main parameters of the model are as follows: the number of weak learners is 150, the maximum depth of each tree is 5, the minimum sample number of non-leaf node partition samples, namely leaf nodes, is 5, the out-of-bag score is used, the random state is set to be 10, the number of features is selected to be the square root of the number of original features, and the others are model default values.
7) And inputting the tested data into the fraud detection model, and acquiring parameters of the fraud detection model to obtain the mobile advertisement fraud detection model.
8) Inputting the input characteristics of the mobile application which is not marked into the mobile advertisement fraud detection model to detect the fraud application.
In the embodiment, the input characteristics of the target mobile application are input into the random forest model to obtain a real number py of 0-1, which represents the probability that the target mobile application is a fraud application. And setting the threshold value tau to be 0.5, if py is larger than tau, the target mobile application is a fraud application, and otherwise, the target mobile application is a normal application.
The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.

Claims (6)

1. A mobile advertisement fraud detection method based on heterogeneous graph embedding is characterized by comprising the following steps:
1) acquiring mobile advertisement log data, and preprocessing the data;
2) extracting incidence relation data of users, applications and advertisements in the mobile advertisement ecosystem, and constructing an authorized heterogeneous graph and a meta template corresponding to the authorized heterogeneous graph
Figure FDA0002986491780000011
Figure FDA0002986491780000012
And
Figure FDA0002986491780000013
respectively represent a category set and a relation category set of edges and satisfy
Figure FDA0002986491780000014
3) Defining meta-paths
Figure FDA0002986491780000015
Setting the number of wandering times n and the longest step length l of each node, traversing the nodes in the weighted heterogeneous graph G, and constructing n weighted random wandering paths S of the nodes vv＝{Sv1,Sv2,...,SvnGet the weighted isomerous graphG element path random walk sequence S;
4) constructing a language model, and learning d-dimensional space dense vector representation X belonging to R in P mobile application nodes in the weighted abnormal graph GP ×dForming an input feature vector;
5) manually marking the mobile application of the training part, and setting a label value of each mobile application according to the information of whether the mobile application is a fraud application; the tag of the fraudulent application is set to 1 and the tag of the non-fraudulent application is set to 0, resulting in PtrainIndividual label data
Figure FDA0002986491780000016
Ptrain＜P，PtrainApplying a total number P eta for the training part, wherein eta is a floating point number which is less than 1 but more than 0, and then combining the floating point number with the corresponding input feature vector in the step 4) to form tested data;
6) constructing a fraud detection model for detecting fraudulent mobile applications;
7) inputting the tested data into a fraud detection model, and acquiring parameters of the fraud detection model to obtain a mobile advertisement fraud detection model;
8) inputting the input characteristics of the mobile application which is not marked into the mobile advertisement fraud detection model to detect the fraud application.
2. The mobile advertising fraud detection method based on heterogeneous graph embedding of claim 1, wherein in step 1), the data preprocessing includes data cleaning and missing value filling; the mobile ad log data contains four attributes: a. unique identification attribute: unique identifiers of users, applications, advertisements; b. the time attribute is as follows: the user uses the application to operate the specific time of the advertisement, and the time is accurate to the second; c. position attribute: identifying a geographic location at which the user is located; d. the device attribute is as follows: the model of the device used by the user, the size of the display screen, the operating system.
3. The method of claim 1, wherein the method comprises detecting fraud in mobile advertisements based on heterogeneous graph embeddingIn the following steps: in step 2), the authorized heterogeneous graph G comprises three types of nodes which are user nodes respectively
Figure FDA0002986491780000021
Application node
Figure FDA0002986491780000022
And advertising node
Figure FDA0002986491780000023
The authorized different composition G comprises three node relations which are respectively used by users
Figure FDA0002986491780000024
User operated advertisement
Figure FDA0002986491780000025
Application display advertisement
Figure FDA0002986491780000026
Its corresponding meta template
Figure FDA0002986491780000027
Indicating that there is a mapping function for any node V ∈ V
Figure FDA0002986491780000028
And there is a mapping function for any connected edge
Figure FDA0002986491780000029
The weight of the edge between two adjacent nodes in the weighted abnormal graph G is determined by the corresponding operation information.
4. The method of claim 1, wherein the mobile advertising fraud detection method based on the heterogeneous graph embedding is characterized in that: in step 3), a random of each node is constructedWandering sequence Sti1, 2., n, the sampling mode of the node is divided into two stages: the initial stage and the subsequent stage represent respectively the walk sequence StiLength of 0 to half element path
Figure FDA00029864917800000210
In the migration stage and length between
Figure FDA00029864917800000211
To the migration phase between the longest migration step l.
5. The method for detecting fraud in mobile advertisement based on embedding of heterogeneous graph according to claim 1, characterized in that: in the step 4), the constructed language model is a Skip-gram model and is accelerated by using a negative sampling mode, and the number of negative samples of negative sampling is fn；
The optimization function of the constructed Skip-gram model is as follows:
Figure FDA0002986491780000031
wherein v istFor given node v heterogeneous context
Figure FDA0002986491780000032
A node in (1); theta is a parameter of the model, Xv,
Figure FDA0002986491780000033
For nodes v and vtA corresponding low-dimensional node vector representation; xaA vector representation of a low dimension of any node in the graph;
finally, the dense vector of the nodes in the graph in the low-dimensional space is expressed as
Figure FDA0002986491780000034
Where d is the vector dimension.
6. The method of claim 1, wherein the mobile advertising fraud detection method based on the heterogeneous graph embedding is characterized in that: in step 6), the constructed fraud detection model is a classifier model, including a traditional machine learning model and a deep learning model.
CN201910417284.7A 2019-05-20 2019-05-20 Mobile advertisement fraud detection method based on heterogeneous graph embedding Active CN110189167B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910417284.7A CN110189167B (en) 2019-05-20 2019-05-20 Mobile advertisement fraud detection method based on heterogeneous graph embedding

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910417284.7A CN110189167B (en) 2019-05-20 2019-05-20 Mobile advertisement fraud detection method based on heterogeneous graph embedding

Publications (2)

Publication Number Publication Date
CN110189167A CN110189167A (en) 2019-08-30
CN110189167B true CN110189167B (en) 2021-06-08

Family

ID=67716841

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910417284.7A Active CN110189167B (en) 2019-05-20 2019-05-20 Mobile advertisement fraud detection method based on heterogeneous graph embedding

Country Status (1)

Country Link
CN (1) CN110189167B (en)

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110660019A (en) * 2019-09-29 2020-01-07 华北电力大学 Small data set simplified stroke generation method based on BPL
CN110958220B (en) * 2019-10-24 2020-12-29 中国科学院信息工程研究所 Network space security threat detection method and system based on heterogeneous graph embedding
CN112988501B (en) * 2019-12-17 2023-02-03 深信服科技股份有限公司 Alarm information generation method and device, electronic equipment and storage medium
CN112995110A (en) * 2019-12-17 2021-06-18 深信服科技股份有限公司 Method and device for acquiring malicious event information and electronic equipment
CN112347260A (en) * 2020-11-24 2021-02-09 深圳市欢太科技有限公司 Data processing method and device and electronic equipment
CN112767054A (en) * 2021-01-29 2021-05-07 北京达佳互联信息技术有限公司 Data recommendation method, device, server and computer-readable storage medium
CN113706279B (en) * 2021-06-02 2024-04-05 同盾科技有限公司 Fraud analysis method, fraud analysis device, electronic equipment and storage medium
CN113553446B (en) * 2021-07-28 2022-05-24 厦门国际银行股份有限公司 Financial anti-fraud method and device based on heterograph deconstruction
CN113656797B (en) * 2021-10-19 2021-12-21 航天宏康智能科技(北京)有限公司 Behavior feature extraction method and behavior feature extraction device
CN114528479B (en) * 2022-01-20 2023-03-21 华南理工大学 Event detection method based on multi-scale heteromorphic image embedding algorithm

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3239919A1 (en) * 2008-03-05 2017-11-01 eBay Inc. Method and apparatus for image recognition services
US8285596B2 (en) * 2008-06-06 2012-10-09 Oracle International Corporation CRM system for enhanced retention of customers
US9818116B2 (en) * 2015-11-11 2017-11-14 Idm Global, Inc. Systems and methods for detecting relations between unknown merchants and merchants with a known connection to fraud
CN109472626B (en) * 2018-11-26 2020-08-18 浙江大学 Intelligent financial risk control method and system for mobile phone leasing service

Also Published As

Publication number Publication date
CN110189167A (en) 2019-08-30

Similar Documents

Publication Publication Date Title
CN110189167B (en) Mobile advertisement fraud detection method based on heterogeneous graph embedding
Li et al. Rhythmic brushstrokes distinguish van Gogh from his contemporaries: findings via automated brushstroke extraction
CN110223106B (en) Deep learning-based fraud application detection method
Li et al. Polygon-based approach for extracting multilane roads from OpenStreetMap urban road networks
CN109726744A (en) A kind of net flow assorted method
CN101334786B (en) Formulae neighborhood based data dimensionality reduction method
JP2019519042A (en) Method and device for pushing information
CN110427990A (en) A kind of art pattern classification method based on convolutional neural networks
CN111008337B (en) Deep attention rumor identification method and device based on ternary characteristics
CN110287329A (en) A kind of electric business classification attribute excavation method based on commodity text classification
CN108763496A (en) A kind of sound state data fusion client segmentation algorithm based on grid and density
CN103530312B (en) Use the method and system of the ID of many-sided footprint
CN113469226B (en) Land utilization classification method and system based on street view image
CN104915354A (en) Multimedia file pushing method and device
CN113364748A (en) Ether house phishing node detection method and system based on transaction subgraph network
CN111680506A (en) External key mapping method and device of database table, electronic equipment and storage medium
CN109783805A (en) A kind of network community user recognition methods and device
Ji et al. Attention based meta path fusion for heterogeneous information network embedding
CN113821702A (en) Urban multidimensional space multivariate heterogeneous information data processing method
CN116955855B (en) Low-cost cross-region address resolution model construction method and system
CN102103700A (en) Land mobile distance-based image spam similarity-detection method
CN117318980A (en) Small sample scene-oriented self-supervision learning malicious traffic detection method
CN106897683A (en) The ground object detecting method and system of a kind of remote sensing images
CN115238773A (en) Malicious account detection method and device for heterogeneous primitive path automatic evaluation
CN111737594B (en) Virtual network role behavior modeling method based on unsupervised label generation

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