CN111767472A - Method and system for detecting abnormal account of social network - Google Patents
Method and system for detecting abnormal account of social network Download PDFInfo
- Publication number
- CN111767472A CN111767472A CN202010649409.1A CN202010649409A CN111767472A CN 111767472 A CN111767472 A CN 111767472A CN 202010649409 A CN202010649409 A CN 202010649409A CN 111767472 A CN111767472 A CN 111767472A
- Authority
- CN
- China
- Prior art keywords
- account
- graph
- model
- encoder
- nodes
- 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
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
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a method and a system for detecting an abnormal account of a social network. The method comprises the following steps: generating a plurality of graphs related to the account graph based on the random block model according to the account graph; determining the structures of an encoder and a decoder, and constructing a graph variation denoising self-encoder model; training the image variation denoising self-encoder model through a plurality of correlation images and initial characteristics, fitting the topological structure of the account image, and finally obtaining the trained image variation denoising self-encoder model; constructing a prediction model; taking an encoder of the graph variation de-noising self-encoder model as a prediction model, and taking parameters of the trained graph variation de-noising self-encoder as initialization parameters of the prediction model; and training a prediction model through the topological structure and the initial characteristics of the account map, and detecting the abnormal account number of the social network. The method and the device can effectively overcome the problem that the number of the nodes with the labels in the graph data is small, so that the abnormal account in the social network can be detected more quickly and accurately.
Description
Technical Field
The invention relates to the field of social networks, in particular to a method and a system for detecting abnormal account numbers of social networks.
Background
The online social network refers to an online network platform for establishing social relationships among people with common backgrounds, interests and behaviors, and with the rapid development of Web2.0, more and more social platforms appear like bamboo shoots in spring after rain, such as microblogs, WeChats, QQQQ and the like which are daily used by people. The appearance of these social tools undoubtedly brings great convenience to our work, life and communication, but at the same time, the huge number of users attracts attackers, and the social tools become a new paradise for attackers to obtain great benefits. An attacker issues a large amount of malicious information such as advertisements and phishing in a social network site by establishing a large amount of false accounts or stealing normal accounts, so that benefits are obtained. The accounts are called abnormal accounts, and because trust relationships exist among users in the social network, malicious information in the accounts is more harmful than malicious information contained in traditional junk mails and the like, and seriously harms a reputation evaluation system of the online social network and trust relationships among the users.
Disclosure of Invention
The invention aims to provide a method and a system for detecting an abnormal account number in a social network, which are used for quickly and accurately detecting the abnormal account number in the social network.
In order to achieve the purpose, the invention provides the following scheme:
a social network abnormal account detection method comprises the following steps:
constructing a social network account graph; the account graph comprises nodes, connecting edges and initial characteristics of the nodes; the node represents a user, and the initial feature represents a user feature;
generating a plurality of graphs related to the account graph based on a random block model according to the account graph;
determining the structures of an encoder and a decoder, and constructing a graph variation denoising self-encoder model;
training the graph variation denoising self-encoder model through a plurality of correlation graphs and the initial characteristics, fitting the topological structure of the account graph, and finally obtaining the trained graph variation denoising self-encoder model;
constructing a prediction model; taking an encoder of the graph variation denoising self-encoder model as a prediction model, and taking parameters of the trained graph variation denoising self-encoder as initialization parameters of the prediction model;
and training the prediction model through the topological structure of the account graph and the initial characteristics, and detecting abnormal account numbers of the social network.
Optionally, the generating a plurality of diagrams related to the account map based on a random block model according to the account map specifically includes:
inputting the account map into the random block model to obtain an output result; the output result comprises a block corresponding to each node, a connection probability matrix among nodes in the block and a connection probability matrix among nodes in the block;
and generating a plurality of correlation graphs according to the connection probability matrix among the nodes in the block and the connection probability matrix among the nodes among the blocks.
Optionally, the training of the graph variation denoising self-encoder model through the multiple correlation graphs and the initial features and the fitting of the topological structure of the account graph are performed to finally obtain the trained graph variation denoising self-encoder model, which specifically includes:
inputting the correlation diagram and the initial characteristics into an encoder of the graph variation de-noising self-encoder model to obtain a distribution matrix represented by each node;
inputting the distribution matrix into a decoder of the graph variation de-noising self-encoder model to obtain a topological structure of the correlation graph;
and fitting the topological structure of the account map according to the topological structure of the correlation map.
Optionally, the method for detecting the social network abnormal account by using the encoder of the graph variation denoising self-encoder model as a prediction model and using the trained graph variation denoising self-encoder parameter as an initialization parameter of the prediction model, training the prediction model, and specifically includes:
taking the trained parameters of the image variational de-noising self-encoder as initialization parameters of a prediction model;
inputting the topological structure of the account map and the initial features into the prediction model, and training the prediction model to obtain a prediction label of each node in the account map;
judging whether the error between the predicted tag and the real tag is within a threshold range;
if yes, obtaining a trained prediction model;
if not, adjusting the parameters of the prediction model to enable the error between the prediction label and the real label to be within the threshold range.
The invention also provides a system for detecting the abnormal account of the social network, which comprises the following steps:
the account map building module is used for building a social network account map; the account graph comprises nodes, connecting edges and initial characteristics of the nodes; the node represents a user, and the initial feature represents a user feature;
a correlation diagram generation module, configured to generate a plurality of diagrams related to the account diagram based on a random block model according to the account diagram;
the first model building module is used for determining the structures of the encoder and the decoder and building a graph variation denoising self-encoder model;
the first training module is used for training the graph variation denoising self-encoder model through a plurality of related graphs and the initial characteristics, fitting the topological structure of the account graph and finally obtaining the trained graph variation denoising self-encoder model;
the second model building module is used for building a prediction model; taking an encoder of the graph variation denoising self-encoder model as a prediction model, and taking parameters of the trained graph variation denoising self-encoder as initialization parameters of the prediction model;
and the second training module is used for training the prediction model through the topological structure of the account map and the initial characteristics to detect the abnormal account of the social network.
Optionally, the correlation diagram generating module specifically includes:
the input unit is used for inputting the account map into the random block model to obtain an output result; the output result comprises a block corresponding to each node, a connection probability matrix among nodes in the block and a connection probability matrix among nodes in the block;
and the generating unit is used for generating a plurality of correlation graphs according to the connection probability matrix among the nodes in the block and the connection probability matrix among the nodes among the blocks.
Optionally, the first training module specifically includes:
the distribution matrix determining unit is used for inputting the correlation diagram and the initial characteristics into an encoder of the graph variation de-noising self-encoder model to obtain a distribution matrix represented by each node;
the correlation diagram topological structure determining unit is used for inputting the distribution matrix to a decoder of the graph variation denoising self-encoder model to obtain the topological structure of the correlation diagram;
and the fitting unit is used for fitting the topological structure of the account map according to the topological structure of the correlation map.
Optionally, the second training module specifically includes:
the prediction label determining unit is used for inputting the topological structure of the account map and the initial features into the prediction model, and training the prediction model to obtain the prediction labels of all nodes in the account map;
the judging unit is used for judging whether the error between the predicted label and the real label is within a threshold value range or not;
the result determining unit is used for obtaining a trained prediction model when the error between the prediction label and the real label is within a threshold range;
an adjusting unit, configured to adjust parameters of the prediction model when an error between the predicted tag and the real tag is outside a threshold range, so that the error between the predicted tag and the real tag is within the threshold range
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
1) the random block model in the invention can generate a plurality of graph data related to the original graph data on the premise of keeping the community structure existing in the real social network, and the graph data can remove redundant connecting edges existing in the original data or increase missing connecting edges in the original graph data. The generated correlation diagram data are used for training the model, so that the model can absorb useful information contained in the data, and the problem of noise in the diagram data is effectively solved.
2) The training image variation denoising self-encoder is an unsupervised learning process, namely label information of nodes is not needed in the training process. The input is subjected to the intermediate representation obtained by the encoder, and the original graph topological structure is restored by the intermediate representation through the decoder, so that the trained encoder has strong representation learning capability, and can learn effective node feature representation which is the key for realizing node classification. The representation capability of the model is characterized by the trained encoder parameters, and the parameters are used as initialization parameters of the prediction model, so that the prediction model has good representation capability at the beginning, and the process of training the prediction model is not a process of training from the beginning according to the labeled data, so that the problem of small number of labeled nodes in the graph data can be effectively solved. The method and the device can detect the abnormal account in the social network more quickly and accurately.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
Fig. 1 is a flowchart of a social network abnormal account detection method according to an embodiment of the present invention;
fig. 2 is a block diagram of a social network abnormal account detection system according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide a method and a system for detecting an abnormal account number in a social network, which are used for quickly and accurately detecting the abnormal account number in the social network.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
As shown in fig. 1, the social network abnormal account detection method includes the following steps:
step 101: constructing a social network account graph; the account graph comprises nodes, connecting edges and initial characteristics of the nodes; the node represents a user and the initial feature represents a user characteristic.
Constructing a social network account graph, known as (V, E, X)v) Wherein V represents the set of all nodes in the graph, E represents the set of all connected edges in the graph, and XvAn initial set of features representing all nodes (note: the features are carried by the nodes themselves, e.g. in a social network, each node represents a user, and the features of the user may contain information such as age, occupation, hobbies, etc., and the features cannot be used directly for node classification tasks.) given the label y of a part of the nodes L (L ∈ V) in the graphLThe target is to predict the label y of the remaining unlabeled node U-V \ L in the graphU。
Step 102: and generating a plurality of graphs related to the account graph based on a random block model according to the account graph.
The random block model (SBM) is an important statistical network analysis model and can effectively process a network without prior knowledge, SILvb is an SBM learning method with model selection capability, a variational EM method is adopted for parameter learning, and Ilvb is utilized for model selection. The invention utilizes the SILvb method to learn on the original graph data (account graph) to obtain the connection probability matrix between the nodes in the blocks and the connection probability matrix between the nodes in the blocks, thereby generating a plurality of graph data related to the original graph data.
The specific flow is that the input of SILvb algorithm is original graph data G, the output is the result of community division of graph G, namely the block to which each node V ∈ V belongs in graph G, the connection probability matrix p between nodes in block and the connection probability matrix q between nodes in blocki(i＝1，2，…，n)。
Step 103: determining the structures of an encoder and a decoder, and constructing a graph variation denoising self-encoder model.
The graph variational self-encoder is a deep neural network model which carries out calculation on graph data, and the model is characterized in that a training process is unsupervised, namely, a data label is not needed. In particular, the model consists of two parts, an encoder and a decoder, both of which are neural networks, the structure can be flexibly set. For the input raw graph data, the encoder maps it into a representation space, the corresponding representation is learned for each node in the graph, unlike the ordinary self-encoder which can only learn the specific representation of the node, the variational self-encoder can learn the distribution of the node representation in the graph, the distribution can depict the uncertainty of the node characteristics, and the distribution can contain more effective information than the deterministic representation. After the node characteristic distribution is obtained, the model samples from the distribution, the original topological structure of the graph is restored through a decoder according to the sampling result, and the model parameters can be updated by using a gradient descent algorithm. Denoising self-encoder means that some noise is added to original data before the original data is sent into an encoder, so that the input of the encoder is noise data, but the original data is still required to be fitted by a specified decoder, and the purpose of doing so is to make a model have stronger robustness and learn characteristics with stronger representation capability. The idea of the denoising self-Encoder is mainly applied to the regular data such as images, texts and the like at present, the invention firstly proposes to introduce the idea into the Graph variant self-Encoder and provides a new Graph variant denoising self-Encoder model (DGVAE).
Step 104: and training the graph variation denoising self-encoder model through a plurality of correlation graphs and the initial characteristics, fitting the topological structure of the account graph, and finally obtaining the trained graph variation denoising self-encoder model.
The Graph variation Denoising self-Encoder model (DGVAE) consists of an Encoder Encoder and a Decoder Decoder, and the task to be solved is a semi-supervised node classification task, and the key point of solving the task is to learn good representation for all nodes in the Graph, so that the nodes can be effectively classified by using the model. The Graph Convolutional neural Network (GCN) is a well-known model of a Graph neural Network, and is a model commonly used for Graph learning, so the present invention takes the GCN as an example and uses the GCN as an encoder model of DGVAE. The GCN consists of two graph convolution layers, and the computation process for each layer can be described as H ═ AXW, where a denotes the adjacency matrix of the graph (a is equivalent to E described above) and X denotes the feature matrix of all nodes in the graph (X and X described above)vEquivalence), W represents a parameter matrix, and H represents a node characteristic updated by one layer of graph convolution layer. Since the object of Decoder is to predict the condition of connecting edges between nodes, the prediction can be performed by calculating the inner product between nodes. After the Encode and Decode structures are determined, the model can be trained, the training process is a repeated iteration process, and the process of only one iteration is described here. In each iteration, the input to the Encoder Encoder is the generated graph GiObtaining a graph G through Encoder calculationiThe distribution matrix Z of all the nodes in the tree is represented; for graph GiFrom distribution Zn(n-1, …, | V |) where sampling one indicates zn(n-1, …, | V |), the sampled representations of all nodes in the graph form a matrix z; restoring graph G by inputting z into DecoderiTopology E ofiAnd finallyWith EiThe topology E of the raw map data G is fitted. At this point, the iterative process is ended, and the process is repeated for a plurality of times until the model converges.
Step 105: constructing a prediction model; and taking an encoder of the graph variation de-noising self-encoder model as a prediction model, and taking parameters of the trained graph variation de-noising self-encoder as initialization parameters of the prediction model.
The trained encoder has strong representation learning capability, and the representation capability of the encoder is completely characterized by the parameters of the encoder. Generally, when training the GCN, the parameters of the model are initialized randomly, which means that the model needs to learn a strong representation ability from the beginning according to the labeled data, and therefore, a large amount of labeled data is needed to train the model. The invention provides that the trained encoder parameters are used as initialization parameters, so that the prediction model has good representation capability at the beginning, and the process of training the prediction model is not the process of training from the beginning according to the labeled data, thereby effectively overcoming the problem of less labeled nodes in the graph data.
Step 106: and training the prediction model through the topological structure of the account graph and the initial characteristics, and detecting abnormal account numbers of the social network.
The input of the prediction model is the topology E of the original graph data G and the initial characteristics X of the nodesvThe output of the prediction model is the prediction labels y of all nodes in the original graphn(n-1, …, | V |), the training process is a supervised process, and the loss function is the cross-entropy loss between the predicted label and the true label. And training the model until convergence by taking the trained Encoder parameter as an initialization parameter of the prediction model, wherein the converged model can predict the node label with unknown label and corresponds to the problem of abnormality detection of the social network, namely whether the user is abnormal or not can be predicted. It should be noted that, since the parameters of the prediction model are no longer initialized randomly, and have strong representation learning ability and contain rich information of the multiple graph data generated in step 102, the model training process needs less than the training process of the original modelThe label can overcome the problem of noise existing in the original image data.
As shown in fig. 2, the present invention further provides a system for detecting an abnormal account in a social network, including:
an account map construction module 201, configured to construct a social network account map; the account graph comprises nodes, connecting edges and initial characteristics of the nodes; the node represents a user and the initial feature represents a user characteristic.
A correlation diagram generating module 202, configured to generate a plurality of diagrams related to the account diagram based on a random block model according to the account diagram.
The correlation diagram generating module 202 specifically includes:
the input unit is used for inputting the account map into the random block model to obtain an output result; the output result comprises a block corresponding to each node, a connection probability matrix among nodes in the block and a connection probability matrix among nodes in the block;
and the generating unit is used for generating a plurality of correlation graphs according to the connection probability matrix among the nodes in the block and the connection probability matrix among the nodes among the blocks.
And the first model building module 203 is used for determining the structures of the encoder and the decoder and building the graph variation denoising self-encoder model.
The first training module 204 is configured to train the graph variation denoising self-encoder model through a plurality of correlation graphs and the initial features, and fit a topological structure of the account graph to finally obtain the trained graph variation denoising self-encoder model.
The first training module 204 specifically includes:
the distribution matrix determining unit is used for inputting the correlation diagram and the initial characteristics into an encoder of the graph variation de-noising self-encoder model to obtain a distribution matrix represented by each node;
the correlation diagram topological structure determining unit is used for inputting the distribution matrix to a decoder of the graph variation denoising self-encoder model to obtain the topological structure of the correlation diagram;
and the fitting unit is used for fitting the topological structure of the account map according to the topological structure of the correlation map.
A second model building module 205, configured to build a prediction model; and taking an encoder of the graph variation de-noising self-encoder model as a prediction model, and taking parameters of the trained graph variation de-noising self-encoder as initialization parameters of the prediction model.
The second training module 206 is configured to train the prediction model according to the topological structure of the account map and the initial features, and detect an abnormal account of a social network.
The second training module 206 specifically includes:
the prediction label determining unit is used for inputting the topological structure of the account map and the initial features into the prediction model, and training the prediction model to obtain the prediction labels of all nodes in the account map;
the judging unit is used for judging whether the error between the predicted label and the real label is within a threshold value range or not;
the result determining unit is used for obtaining a trained prediction model when the error between the prediction label and the real label is within a threshold range;
and the adjusting unit is used for adjusting the parameters of the prediction model when the error between the predicted tag and the real tag is out of the threshold range, so that the error between the predicted tag and the real tag is in the threshold range.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (8)
1. A social network abnormal account detection method is characterized by comprising the following steps:
constructing a social network account graph; the account graph comprises nodes, connecting edges and initial characteristics of the nodes; the node represents a user, and the initial feature represents a user feature;
generating a plurality of graphs related to the account graph based on a random block model according to the account graph;
determining the structures of an encoder and a decoder, and constructing a graph variation denoising self-encoder model;
training the graph variation denoising self-encoder model through a plurality of correlation graphs and the initial characteristics, fitting the topological structure of the account graph, and finally obtaining the trained graph variation denoising self-encoder model;
constructing a prediction model; taking an encoder of the graph variation denoising self-encoder model as a prediction model, and taking parameters of the trained graph variation denoising self-encoder as initialization parameters of the prediction model;
and training the prediction model through the topological structure of the account graph and the initial characteristics, and detecting abnormal account numbers of the social network.
2. The social network abnormal account detection method according to claim 1, wherein the generating a plurality of graphs related to the account graph based on a random block model according to the account graph specifically includes:
inputting the account map into the random block model to obtain an output result; the output result comprises a block corresponding to each node, a connection probability matrix among nodes in the block and a connection probability matrix among nodes in the block;
and generating a plurality of correlation graphs according to the connection probability matrix among the nodes in the block and the connection probability matrix among the nodes among the blocks.
3. The social network abnormal account detection method according to claim 1, wherein the training of the graph variation denoising self-encoder model through the plurality of correlation graphs and the initial features is performed to fit a topological structure of the account graph, so as to finally obtain the trained graph variation denoising self-encoder model, and specifically includes:
inputting the correlation diagram and the initial characteristics into an encoder of the graph variation de-noising self-encoder model to obtain a distribution matrix represented by each node;
inputting the distribution matrix into a decoder of the graph variation de-noising self-encoder model to obtain a topological structure of the correlation graph;
and fitting the topological structure of the account map according to the topological structure of the correlation map.
4. The method according to claim 1, wherein the training of the prediction model through the topological structure of the account map and the initial features to detect the abnormal social network account specifically comprises:
inputting the topological structure of the account map and the initial features into the prediction model, and training the prediction model to obtain a prediction label of each node in the account map;
judging whether the error between the predicted tag and the real tag is within a threshold range;
if yes, obtaining a trained prediction model;
if not, adjusting the parameters of the prediction model to enable the error between the prediction label and the real label to be within the threshold range.
5. A social network abnormal account detection system is characterized by comprising:
the account map building module is used for building a social network account map; the account graph comprises nodes, connecting edges and initial characteristics of the nodes; the node represents a user, and the initial feature represents a user feature;
a correlation diagram generation module, configured to generate a plurality of diagrams related to the account diagram based on a random block model according to the account diagram;
the first model building module is used for determining the structures of the encoder and the decoder and building a graph variation denoising self-encoder model;
the first training module is used for training the graph variation denoising self-encoder model through a plurality of related graphs and the initial characteristics, fitting the topological structure of the account graph and finally obtaining the trained graph variation denoising self-encoder model;
the second model building module is used for building a prediction model; taking an encoder of the graph variation denoising self-encoder model as a prediction model, and taking parameters of the trained graph variation denoising self-encoder as initialization parameters of the prediction model;
and the second training module is used for training the prediction model through the topological structure of the account map and the initial characteristics to detect the abnormal account of the social network.
6. The social network abnormal account detection system according to claim 5, wherein the correlation diagram generation module specifically includes:
the input unit is used for inputting the account map into the random block model to obtain an output result; the output result comprises a block corresponding to each node, a connection probability matrix among nodes in the block and a connection probability matrix among nodes in the block;
and the generating unit is used for generating a plurality of correlation graphs according to the connection probability matrix among the nodes in the block and the connection probability matrix among the nodes among the blocks.
7. The social network abnormal account detection system according to claim 5, wherein the first training module specifically includes:
the distribution matrix determining unit is used for inputting the correlation diagram and the initial characteristics into an encoder of the graph variation de-noising self-encoder model to obtain a distribution matrix represented by each node;
the correlation diagram topological structure determining unit is used for inputting the distribution matrix to a decoder of the graph variation denoising self-encoder model to obtain the topological structure of the correlation diagram;
and the fitting unit is used for fitting the topological structure of the account map according to the topological structure of the correlation map.
8. The social network abnormal account detection system according to claim 5, wherein the second training module specifically comprises:
the prediction label determining unit is used for inputting the topological structure of the account map and the initial features into the prediction model, and training the prediction model to obtain the prediction labels of all nodes in the account map;
the judging unit is used for judging whether the error between the predicted label and the real label is within a threshold value range or not;
the result determining unit is used for obtaining a trained prediction model when the error between the prediction label and the real label is within a threshold range;
and the adjusting unit is used for adjusting the parameters of the prediction model when the error between the predicted tag and the real tag is out of the threshold range, so that the error between the predicted tag and the real tag is in the threshold range.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010649409.1A CN111767472A (en) | 2020-07-08 | 2020-07-08 | Method and system for detecting abnormal account of social network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010649409.1A CN111767472A (en) | 2020-07-08 | 2020-07-08 | Method and system for detecting abnormal account of social network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111767472A true CN111767472A (en) | 2020-10-13 |
Family
ID=72725185
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010649409.1A Pending CN111767472A (en) | 2020-07-08 | 2020-07-08 | Method and system for detecting abnormal account of social network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111767472A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112445957A (en) * | 2020-11-05 | 2021-03-05 | 西安电子科技大学 | Social network abnormal user detection method, system, medium, equipment and terminal |
| CN112818257A (en) * | 2021-02-19 | 2021-05-18 | 北京邮电大学 | Account detection method, device and equipment based on graph neural network |
| CN113158076A (en) * | 2021-04-05 | 2021-07-23 | 北京工业大学 | Social robot detection method based on variational self-coding and K-nearest neighbor combination |
| CN113420148A (en) * | 2021-06-22 | 2021-09-21 | 中国平安人寿保险股份有限公司 | Training text acquisition method, system and equipment for sensitive content quality inspection model |
| CN113537272A (en) * | 2021-03-29 | 2021-10-22 | 之江实验室 | Semi-supervised social network abnormal account detection method based on deep learning |
| CN113609306A (en) * | 2021-08-04 | 2021-11-05 | 北京邮电大学 | Social network link prediction method and system for resisting residual image variation self-encoder |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108009220A (en) * | 2017-11-22 | 2018-05-08 | 中国电子科技集团公司第二十八研究所 | A kind of method for being detected in network hotspot public sentiment event and positioning abnormal user |
| CN108985330A (en) * | 2018-06-13 | 2018-12-11 | 华中科技大学 | A kind of autoencoder network and its training method abnormal use electric detection method and system |
| CN109905399A (en) * | 2019-03-14 | 2019-06-18 | 哈尔滨工程大学 | A kind of social media individual abnormal user detection method to be developed based on self network structure |
| CN110191110A (en) * | 2019-05-20 | 2019-08-30 | 山西大学 | Social networks exception account detection method and system based on network representation study |
| CN110781406A (en) * | 2019-10-14 | 2020-02-11 | 西安交通大学 | Social network user multi-attribute inference method based on variational automatic encoder |
| CN111291190A (en) * | 2020-03-23 | 2020-06-16 | 腾讯科技（深圳）有限公司 | Training method of encoder, information detection method and related device |
-
2020
- 2020-07-08 CN CN202010649409.1A patent/CN111767472A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108009220A (en) * | 2017-11-22 | 2018-05-08 | 中国电子科技集团公司第二十八研究所 | A kind of method for being detected in network hotspot public sentiment event and positioning abnormal user |
| CN108985330A (en) * | 2018-06-13 | 2018-12-11 | 华中科技大学 | A kind of autoencoder network and its training method abnormal use electric detection method and system |
| CN109905399A (en) * | 2019-03-14 | 2019-06-18 | 哈尔滨工程大学 | A kind of social media individual abnormal user detection method to be developed based on self network structure |
| CN110191110A (en) * | 2019-05-20 | 2019-08-30 | 山西大学 | Social networks exception account detection method and system based on network representation study |
| CN110781406A (en) * | 2019-10-14 | 2020-02-11 | 西安交通大学 | Social network user multi-attribute inference method based on variational automatic encoder |
| CN111291190A (en) * | 2020-03-23 | 2020-06-16 | 腾讯科技（深圳）有限公司 | Training method of encoder, information detection method and related device |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112445957A (en) * | 2020-11-05 | 2021-03-05 | 西安电子科技大学 | Social network abnormal user detection method, system, medium, equipment and terminal |
| CN112818257A (en) * | 2021-02-19 | 2021-05-18 | 北京邮电大学 | Account detection method, device and equipment based on graph neural network |
| CN113537272A (en) * | 2021-03-29 | 2021-10-22 | 之江实验室 | Semi-supervised social network abnormal account detection method based on deep learning |
| CN113537272B (en) * | 2021-03-29 | 2024-03-19 | 之江实验室 | Deep learning-based semi-supervised social network abnormal account detection method |
| CN113158076A (en) * | 2021-04-05 | 2021-07-23 | 北京工业大学 | Social robot detection method based on variational self-coding and K-nearest neighbor combination |
| CN113158076B (en) * | 2021-04-05 | 2022-07-22 | 北京工业大学 | Social robot detection method based on variational self-coding and K-nearest neighbor combination |
| CN113420148A (en) * | 2021-06-22 | 2021-09-21 | 中国平安人寿保险股份有限公司 | Training text acquisition method, system and equipment for sensitive content quality inspection model |
| CN113420148B (en) * | 2021-06-22 | 2024-02-09 | 中国平安人寿保险股份有限公司 | Training text acquisition method, system and equipment for sensitive content quality inspection model |
| CN113609306A (en) * | 2021-08-04 | 2021-11-05 | 北京邮电大学 | Social network link prediction method and system for resisting residual image variation self-encoder |
| CN113609306B (en) * | 2021-08-04 | 2024-04-23 | 北京邮电大学 | Social network link prediction method and system for anti-residual diagram variation self-encoder |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111767472A (en) | Method and system for detecting abnormal account of social network | |
| CN112818257B (en) | Account detection method, device and equipment based on graph neural network | |
| CN110933105B (en) | Web attack detection method, system, medium and equipment | |
| CN113961759B (en) | Abnormality detection method based on attribute map representation learning | |
| CN112231562A (en) | Network rumor identification method and system | |
| CN113407784B (en) | Social network-based community dividing method, system and storage medium | |
| Opolka et al. | Spatio-temporal deep graph infomax | |
| CN114781609A (en) | Traffic flow prediction method based on multi-mode dynamic residual image convolution network | |
| CN114817663A (en) | Service modeling and recommendation method based on class perception graph neural network | |
| CN112507245B (en) | Social network friend recommendation method based on graph neural network | |
| CN113783874A (en) | Network security situation assessment method and system based on security knowledge graph | |
| CN112667920A (en) | Text perception-based social influence prediction method, device and equipment | |
| CN109657725B (en) | Service quality prediction method and system based on complex space-time context awareness | |
| CN113541986B (en) | Fault prediction method and device for 5G slice and computing equipment | |
| Gao et al. | The prediction role of hidden markov model in intrusion detection | |
| CN117272195A (en) | Block chain abnormal node detection method and system based on graph convolution attention network | |
| CN116127325A (en) | Method and system for detecting abnormal flow of graph neural network business based on multi-attribute graph | |
| Hu et al. | An Adaptive Filtering Algorithm Based on Genetic Algorithm‐Backpropagation Network | |
| CN113343041B (en) | Message reply relation judgment system based on graph model representation learning | |
| CN113343235B (en) | Application layer malicious effective load detection method, system, device and medium based on Transformer | |
| CN115102868A (en) | Web service QoS prediction method based on SOM clustering and depth self-encoder | |
| CN114866246A (en) | Computer network security intrusion detection method based on big data | |
| Yang et al. | Deep Learning‐Based Destination Prediction Scheme by Trajectory Prediction Framework | |
| CN117113240B (en) | Dynamic network community discovery method, device, equipment and storage medium | |
| CN115982508B (en) | Heterogeneous information network-based website detection method, electronic equipment and medium |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20201013 |
|
| RJ01 | Rejection of invention patent application after publication |