Anti-fraud method, system and device based on behavior characteristics of old people and storage medium

Abstract

The invention discloses an anti-fraud method, system and device based on behavior characteristics of old people and a storage medium. The method comprises the following steps: constructing a twin network; training a twin network by using historical communication voice data of the old; taking the call records as nodes, calculating similarities among the nodes, and taking the similarities as edges among the nodes to construct a call behavior relation graph; performing clustering analysis on the nodes in the call behavior relation graph based on the similarity between the nodes to obtain an outlier degree of the nodes, and if the outlier degree is higher than a set threshold value, adding a mark to a number corresponding to the node once; and obtaining the marked frequencies of all numbers in the database recently, and if the marked frequencies are higher than a set threshold value, determining that the numbers are suspicious numbers.

Description

Technical Field

[0001] The present invention relates to the field of fraudulent phone call identification, and in particular to an anti-fraud method, system, device and storage medium based on elderly behavioral characteristics. Background Art

[0003] Most existing methods extract features based on the caller's communication data and compare them with the features of fraudulent calls to analyze the suspiciousness of the caller. On the one hand, the existing methods focus on a wide range of victims.

[0004] On the other hand, existing methods usually assume that the fraud call is the caller and therefore focus more on the caller's communication data. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides an anti-fraud method, system, device and storage medium based on the behavioral characteristics of the elderly.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] An anti-fraud method based on elderly behavioral characteristics comprises the following steps:

[0008] Build a twin network based on LSTM module;

[0009] Use the historical communication voice data of the elderly to train the twin network and obtain a trained twin network;

[0010] Take call records as nodes, calculate the similarity between nodes, use the similarity as the edge between nodes, and build a call behavior relationship graph;

[0011] The similarity between the computing nodes is specifically:

[0012] Extract the behavior feature vector of the call object corresponding to the call record; use the trained twin network to extract the voice feature vector of the call voice corresponding to the call record;

[0013] The behavior feature vector of each node is concatenated with the corresponding speech feature vector, and each node obtains a concatenated feature vector; the similarity between the concatenated feature vectors is used as the similarity between the nodes;

[0014] Perform cluster analysis on the nodes in the call behavior relationship graph based on the similarity between the nodes to obtain the outlier degree of the node. If the outlier degree is higher than the set threshold, add a mark to the number corresponding to the node;

[0015] Obtain the marked frequency of all numbers in the recent database. If the marked frequency is higher than the set threshold, it is considered a suspicious number and the elderly are reminded.

[0016] To optimize the above technical solutions, the specific measures taken also include:

[0017] Furthermore, the twin network includes a first LSTM subnetwork and a second LSTM subnetwork, the first LSTM subnetwork and the second LSTM subnetwork do not share parameters, the first LSTM subnetwork is used to reduce the dimension of the incoming call voice sequence, and the second LSTM subnetwork is used to reduce the dimension of the outgoing call voice sequence.

[0018] Furthermore, the historical communication voice data includes the communication voice and a tag for marking the topic type;

[0019] The specific method of training the twin network using the historical communication voice data of the elderly is as follows:

[0020] Randomly select two voices from historical communication voice data. The selection method is: if the topics are the same, select a label from multiple labels in order, and randomly select two voices from the communication voices marked with the label to form a data pair; if the topics are different, select a label from multiple labels in order, randomly select a voice from the data marked with the label, and randomly select another voice from the data not marked with the label, and form the two voices into a data pair;

[0021] According to the types of the two voices, it is divided into the following 6 situations:

[0022] a. Both are incoming calls with the same topic;

[0023] b. Both are incoming calls, and the content of the calls do not have the same topic;

[0024] c. Both are outgoing voice calls with the same topic;

[0025] d. Both are outgoing voice calls, and the content of the voice calls do not have the same topic;

[0026] e. One is an incoming call and the other is an outgoing call, and the voice content is on the same topic;

[0027] f. One is for incoming calls, the other is for outgoing calls, and the content of the voices do not have the same topic

[0028] Select one of the above 6 situations from the historical communication voice data in turn to train the first LSTM sub-network and the second LSTM sub-network;

[0029] The objective function is designed so that the twin network can distinguish the similarities of different speech during training.

[0030] Furthermore, the objective function is specifically:

[0031] L(x1,y1,x2,y2,y3)=f3*D(x1,f1,x2,f2)+(1-y3)*max(bD(x1,y1,x2,y2),0)+R(y1,y2)

[0032] Wherein, b is the preset boundary value, L(x1,y1,x2,y2,y3) represents the objective function, x1 represents the first speech segment, y1 represents whether the first speech segment is an incoming call or an outgoing call, x2 represents the second speech segment, y2 represents whether the second speech segment is an incoming call or an outgoing call, y3 represents whether the two speech segments have the same topic, D(x1,y1,x2,y2) represents the distance between the two speech segments in the vector space, max(·) represents taking the maximum value, and R(y1,y2) represents the regularization term;

[0033] in:

[0034] D(x1,y1,x2,y2)=(1-y1)*(1-y2)*((M A (x1)-M A (x2)) 2 +(y1+y2-2*y1*y2)*(M A (x1)-M B (x2)) 2 +y1*y2*(M B (x v )-M B (x2)) 2

[0035] R(y1,y2)=(1-y1*y2)*M A 2 +(y1+y2-y1*y2)*M B 2

[0036] Where M A (x1) represents the vector after the first LSTM sub-network reduces the dimension of the first speech segment, M A (x2) represents the vector after the first LSTM sub-network reduces the dimension of the second speech segment, M B (x1) represents the vector after the second LSTM sub-network reduces the dimension of the first speech segment, M B (x2) represents the vector after the second LSTM sub-network reduces the dimension of the second speech segment, M A represents the parameters of the first LSTM sub-network, M B Represents the parameters of the second LSTM sub-network;

[0037] in:

[0038]

[0039] Furthermore, the behavior feature vector of the call object corresponding to the call record is extracted as follows:

[0040] For each call object, we divide it into incoming and outgoing calls, and extract the recent distribution of call behavior, including call frequency, call time period, call duration, latitude and longitude of the call object, age of the call object, and network access time, as the behavior feature vector f g ;

[0041] The method of using the trained twin network to extract the speech feature vector of the call speech corresponding to the call record is as follows:

[0042] For each call speech, we divide it into incoming and outgoing calls. We use the first LSTM sub-network and the second LSTM sub-network to reduce the dimension of the incoming call speech sequence and the outgoing call speech sequence, respectively, to obtain the speech feature vector f v ;

[0043] The concatenated feature vector is f c =(f g ,f v );

[0044] The similarity between the concatenated feature vectors is specifically measured using the Euclidean distance;

[0045] Before calculating the similarity between the concatenated feature vectors, a process of normalizing the concatenated feature vectors is also included.

[0046] Furthermore, cluster analysis is performed on the nodes in the call behavior relationship graph based on the similarity between the nodes, and the outlier degree of the node is obtained as follows:

[0047] The outlier degree O of node k k Calculation method:

[0048]

[0049] Where u, v, and k represent three different nodes, G represents a node set, e is a natural constant, S(k,v) represents the Euclidean distance between k and v, and S(u,v) represents the Euclidean distance between u and v.

[0050] Furthermore, the marked frequencies of all the numbers are measured by the suspiciousness, and the calculation formula of the suspiciousness is:

[0051]

[0052] In the formula, V represents the suspiciousness, L represents the number of times the number is marked, and N represents the number of recent calls made by the number.

[0053] The present invention also proposes an anti-fraud system based on elderly behavioral characteristics, comprising:

[0054] Network building module, used to build a twin network based on the LSTM module;

[0055] The network training module is used to train the twin network using the historical communication voice data of the elderly to obtain a trained twin network;

[0056] A relationship graph construction module is used to calculate the similarity between nodes using call records as nodes, and use the similarity as the edge between nodes to construct a call behavior relationship graph;

[0057] The similarity between the computing nodes is specifically:

[0058] Extract the behavior feature vector of the call object corresponding to the call record; use the trained twin network to extract the voice feature vector of the call voice corresponding to the call record;

[0059] The behavior feature vector of each node is concatenated with the corresponding speech feature vector, and each node obtains a concatenated feature vector; the similarity between the concatenated feature vectors is used as the similarity between the nodes;

[0060] The number marking module is used to perform cluster analysis on the nodes in the call behavior relationship graph based on the similarity between the nodes to obtain the outlier degree of the node. If the outlier degree is higher than the set threshold, the number corresponding to the node is marked once;

[0061] The reminder module is used to obtain the marked frequency of all numbers in the recent database. If the marked frequency is higher than the set threshold, it is considered to be a suspicious number and the elderly are reminded.

[0062] The present invention also proposes an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the anti-fraud method based on the behavioral characteristics of the elderly as described above is implemented.

[0063] The present invention also proposes a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the anti-fraud method based on the behavioral characteristics of the elderly as described above.

[0064] The beneficial effects of the present invention are:

[0065] 1. Most existing fraud call identification methods are based on the analysis of the fraud behavior itself, and generally assume that the fraud call is an incoming call. Therefore, the existing methods have the limitation of being unable to identify phishing frauds, while the present invention has nothing to do with whether the suspicious number is an incoming number.

[0066] 2. Starting from the users themselves, the behavior of the elderly is generally relatively fixed. From this point of view, the behavior of the elderly is analyzed. If abnormal behavior occurs, the number is judged as a suspicious number. At the same time, the fault tolerance of the method is improved by the automatic marking method of multiple users, reducing the probability of misjudgment. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is a flow chart of the anti-fraud method based on the behavioral characteristics of the elderly proposed by the present invention.

[0068] Figure 2 A structural diagram of an electronic device. DETAILED DESCRIPTION

[0069] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0070] Embodiment 1

[0071] The present invention proposes an anti-fraud method based on the behavioral characteristics of the elderly. The overall process is as follows: Figure 1 As shown, the following steps are included:

[0072] A Siamese-LSTM network is constructed based on the LSTM module; the Siamese-LSTM network includes a first LSTM subnetwork and a second LSTM subnetwork, the first LSTM subnetwork and the second LSTM subnetwork do not share parameters, the first LSTM subnetwork is used to reduce the dimension of the incoming call voice sequence, and the second LSTM subnetwork is used to reduce the dimension of the outgoing call voice sequence.

[0073] The historical communication voice data of the elderly is used to train the twin network, so that it can produce a high response to two different voices on the same topic and a low response to two different voices on different topics, and a trained twin network is obtained; the historical communication voice data includes the communication voice and labels used to mark the topic type; such as elderly care, retirement, health, children, etc. The specific steps of using the historical communication voice data of the elderly to train the twin network are:

[0074] Randomly select two voices from historical communication voice data. The selection method is: if the topics are the same, select a label from multiple labels in order, and randomly select two voices from the communication voices marked with the label to form a data pair; if the topics are different, select a label from multiple labels in order, randomly select a voice from the data marked with the label, and randomly select another voice from the data not marked with the label, and form the two voices into a data pair;

[0075] According to the types of the two voices, it is divided into the following 6 situations:

[0076] a. Both are incoming calls with the same topic;

[0077] b. Both are incoming calls, and the content of the calls do not have the same topic;

[0078] c. Both are outgoing voice calls with the same topic;

[0079] d. Both are outgoing voice calls, and the content of the voice calls do not have the same topic;

[0080] e. One is an incoming call and the other is an outgoing call, and the voice content is on the same topic;

[0081] f. One is for incoming calls, the other is for outgoing calls, and the content of the voices do not have the same topic

[0082] Select one of the above 6 situations from the historical communication voice data in turn to train the first LSTM sub-network and the second LSTM sub-network;

[0083] The objective function is designed so that the twin network can distinguish the similarities of different speech during training.

[0084] The specific objective function is:

[0085] L(x1,y1,x2,y2,y3)=y3*D(x1,y1,x2,y2)+(1-y3)*max(bD(x1,y1,x2,y2),0)+R(y1,y2)

[0086] Wherein, b is the preset boundary value, L(x1,y1,x2,y2,y3) represents the objective function, x1 represents the first speech segment, y1 represents whether the first speech segment is an incoming call or an outgoing call, x2 represents the second speech segment, y2 represents whether the second speech segment is an incoming call or an outgoing call, y3 represents whether the two speech segments have the same topic, D(x1,y1,x2,y2) represents the distance between the two speech segments in the vector space, max(·) represents taking the maximum value, and R(y1,y2) represents the regularization term;

[0087] in:

[0088] D(x1,y1,x2,y2)=(1-y1)*(1-y2)*((M A (x1)-M A (x2)) 2 +(y1+y2-2*y1*y2)*(MA (x1)-M B (x2)) 2 +y1*y2*(M1(x1)-M B (x2)) 2

[0089] R(y1,y2)=(1-y1*y2)*M A 2 +(y1+y2-y1*y2)*M B 2

[0090] Where M A (x1) represents the vector after the first LSTM sub-network reduces the dimension of the first speech segment, M A (x2) represents the vector after the first LSTM sub-network reduces the dimension of the second speech segment, M B (x1) represents the vector after the second LSTM sub-network reduces the dimension of the first speech segment, M B (x2) represents the vector after the second LSTM sub-network reduces the dimension of the second speech segment, M A represents the parameters of the first LSTM sub-network, M B Represents the parameters of the second LSTM sub-network;

[0091] in:

[0092]

[0093] Take call records as nodes, calculate the similarity between nodes, use the similarity as the edge between nodes, and build a call behavior relationship graph;

[0094] The similarity between nodes is calculated as follows:

[0095] Extract the behavior feature vector of the call object corresponding to the call record; specifically:

[0096] For each call object, we divide it into incoming and outgoing calls, and extract the recent distribution of call behavior, including call frequency, call time period, call duration, latitude and longitude of the call object, age of the call object, and network access time, as the behavior feature vector f g ;

[0097] The trained twin network is used to extract the speech feature vector of the call speech corresponding to the call record; specifically, for each call speech, it is divided into incoming and outgoing calls, and the first LSTM sub-network and the second LSTM sub-network are used to reduce the dimension of the incoming call speech sequence and the outgoing call speech sequence respectively to obtain the speech feature vector f v ;

[0098] Concatenate the behavior feature vector of each node with the corresponding speech feature vector, and the concatenated feature vector is f c =(f g ,f v ); each node gets a concatenated feature vector; the similarity between the concatenated feature vectors is used as the similarity between the nodes; the similarity between the concatenated feature vectors is specifically measured by the Euclidean distance;

[0099] Before calculating the similarity between the concatenated feature vectors, the concatenated feature vectors are normalized. The normalization formula is as follows:

[0100] In the formula, f s represents the normalized concatenated feature vector, mean(·) represents the mean, and std(·) represents the standard deviation.

[0101] Based on the similarity between nodes, cluster analysis is performed on the nodes in the call behavior relationship graph to obtain the outlier degree of the node. The outlier degree O of node k is k Calculation method:

[0102]

[0103] Where u, v, and k represent three different nodes, G represents a node set, e is a natural constant, S(k,v) represents the Euclidean distance between k and v, and S(u,v) represents the Euclidean distance between u and v.

[0104] If the outlier degree is higher than the set threshold T o , then the number corresponding to the node is marked once; it is marked as an outlier and uploaded to the server after deduplication.

[0105] Get the marked frequency of all numbers in the recent database. If the marked frequency is higher than the set threshold T v , it is considered a suspicious number and the elderly are reminded. Given a threshold T V Belongs to (0,1).

[0106] The frequency of all numbers being marked is measured by the degree of suspicion, and the calculation formula for the degree of suspicion is:

[0107]

[0108] In the formula, V represents the suspiciousness, L represents the number of times the number is marked, and N represents the number of recent calls made by the number.

[0109] Embodiment 2

[0110] The present invention proposes an anti-fraud system based on elderly behavior characteristics corresponding to the method of embodiment 1, comprising:

[0111] Network building module, used to build a twin network based on the LSTM module;

[0112] The network training module is used to train the twin network using the historical communication voice data of the elderly to obtain a trained twin network;

[0113] A relationship graph construction module is used to calculate the similarity between nodes using call records as nodes, and use the similarity as the edge between nodes to construct a call behavior relationship graph;

[0114] The similarity between the computing nodes is specifically:

[0115] Extract the behavior feature vector of the call object corresponding to the call record; use the trained twin network to extract the voice feature vector of the call voice corresponding to the call record;

[0116] The behavior feature vector of each node is concatenated with the corresponding speech feature vector, and each node obtains a concatenated feature vector; the similarity between the concatenated feature vectors is used as the similarity between the nodes;

[0117] The number marking module is used to perform cluster analysis on the nodes in the call behavior relationship graph based on the similarity between the nodes to obtain the outlier degree of the node. If the outlier degree is higher than the set threshold, the number corresponding to the node is marked once;

[0118] The reminder module is used to obtain the marked frequency of all numbers in the recent database. If the marked frequency is higher than the set threshold, it is considered to be a suspicious number and the elderly are reminded.

[0119] The implementation methods of each module and module function in the system are completely consistent with the steps of the method in Example 1, so they will not be repeated here.

[0120] Embodiment 3

[0121] The present invention provides an electronic device, such as Figure 2 As shown, it includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the anti-fraud method based on the behavioral characteristics of the elderly as described in Example 1 is implemented.

[0122] Embodiment 4

[0123] The present invention provides a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the anti-fraud method based on elderly behavioral characteristics as described in Example 1.

[0124] In the embodiments disclosed in the present application, the computer storage medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. The computer storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the above. More specific examples of computer storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0125] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0126] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. An anti-fraud method based on the behavioral characteristics of the elderly, characterized in that: The following steps are involved: Build a twin network based on LSTM module; Use the historical communication voice data of the elderly to train the twin network and obtain a trained twin network; Take call records as nodes, calculate the similarity between nodes, use the similarity as the edge between nodes, and build a call behavior relationship graph; The similarity between the computing nodes is specifically: Extract the behavior feature vector of the call object corresponding to the call record; use the trained twin network to extract the voice feature vector of the call voice corresponding to the call record; Concatenate the behavior feature vector of each node with the corresponding speech feature vector, and obtain a concatenated feature vector for each node; The similarity between the concatenated feature vectors is used as the similarity between the nodes; Perform cluster analysis on the nodes in the call behavior relationship graph based on the similarity between the nodes to obtain the outlier degree of the node. If the outlier degree is higher than the set threshold, add a mark to the number corresponding to the node; Obtain the marked frequency of all numbers in the recent database. If the marked frequency is higher than the set threshold, it is considered a suspicious number and the elderly are reminded.

2. The anti-fraud method based on the behavioral characteristics of the elderly as claimed in claim 1, characterized in that: The twin network includes a first LSTM subnetwork and a second LSTM subnetwork, the first LSTM subnetwork and the second LSTM subnetwork do not share parameters, the first LSTM subnetwork is used to reduce the dimension of the incoming call voice sequence, and the second LSTM subnetwork is used to reduce the dimension of the outgoing call voice sequence.

3. The anti-fraud method based on elderly behavioral characteristics as claimed in claim 1, characterized in that: The historical communication voice data includes the communication voice and a tag for marking the topic type; The specific method of training the twin network using the historical communication voice data of the elderly is as follows: Randomly select two voices from historical communication voice data. The selection method is: if the topics are the same, select a label from multiple labels in order, and randomly select two voices from the communication voices marked with the label to form a data pair; if the topics are different, select a label from multiple labels in order, randomly select a voice from the data marked with the label, and randomly select another voice from the data not marked with the label, and form the two voices into a data pair; According to the types of the two voices, it is divided into the following 6 situations: a. Both are incoming calls with the same topic; b. Both are incoming calls, and the content of the calls do not have the same topic; c. Both are outgoing voice calls with the same topic; d. Both are outgoing voice calls, and the content of the voice calls do not have the same topic; e. One is an incoming call and the other is an outgoing call, and the voice content is on the same topic; f. One is for incoming calls, the other is for outgoing calls, and the content of the voices do not have the same topic Select one of the above 6 situations from the historical communication voice data in turn to train the first LSTM sub-network and the second LSTM sub-network; The objective function is designed so that the twin network can distinguish the similarities of different speech during training.

4. The anti-fraud method based on elderly behavioral characteristics as claimed in claim 3, characterized in that: The objective function is specifically: L(x1,y1,x2,y2,y3)=y3*D(x1,y1,x2,y2)+(1-y3)*max(bD(x1,y1,x2,y2),0)+R(y1,y2) Wherein, b is the preset boundary value, L(x1,y1,x2,y2,y3) represents the objective function, x1 represents the first speech segment, y1 represents whether the first speech segment is an incoming call or an outgoing call, x2 represents the second speech segment, y2 represents whether the second speech segment is an incoming call or an outgoing call, y3 represents whether the two speech segments have the same topic, D(x1,y1,x2,y2) represents the distance between the two speech segments in the vector space, max(·) represents taking the maximum value, and R(y1,y2) represents the regularization term; in: D(x1,y1,x2,y2)=(1-y1)*(1-y2)*((M A (x1)-M A (x2)) 2 +(y1+y2-2*y1-y2)*(M A (x1)-M B (x2)) 2 +y1*y2*(M B (x1)-M B (x2)) 2 R(y1,y2)=(1-y1*y2)*M A 2 +(y1+y2-y1*y2)*M B 2 Where M A (x1) represents the vector after the first LSTM sub-network reduces the dimension of the first speech segment, M A (x2) represents the vector after the first LSTM sub-network reduces the dimension of the second speech segment, M B (x1) represents the vector after the second LSTM sub-network reduces the dimension of the first speech segment, M B (x2) represents the vector after the second LSTM sub-network reduces the dimension of the second speech segment, M A represents the parameters of the first LSTM sub-network, M B Represents the parameters of the second LSTM sub-network; in:

5. The anti-fraud method based on elderly behavioral characteristics as claimed in claim 1, characterized in that: The behavior feature vector of the call object corresponding to the call record is specifically: For each call object, we divide it into incoming and outgoing calls, and extract the recent distribution of call behavior, including call frequency, call time period, call duration, latitude and longitude of the call object, age of the call object, and network access time, as the behavior feature vector f g ; The method of using the trained twin network to extract the speech feature vector of the call speech corresponding to the call record is specifically as follows: For each call speech, we divide it into incoming and outgoing calls. We use the first LSTM sub-network and the second LSTM sub-network to reduce the dimension of the incoming call speech sequence and the outgoing call speech sequence, respectively, to obtain the speech feature vector f v ; The concatenated feature vector is f c =(f g ,f v ); The similarity between the concatenated feature vectors is specifically measured using the Euclidean distance; Before calculating the similarity between the concatenated feature vectors, a process of normalizing the concatenated feature vectors is also included.

6. The anti-fraud method based on elderly behavioral characteristics as claimed in claim 1, characterized in that: The nodes in the call behavior relationship graph are clustered based on the similarity between the nodes, and the outlier degree of the node is obtained as follows: The outlier degree O of node k k Calculation method: Where u, v, and k represent three different nodes, G represents a node set, e is a natural constant, S(k,v) represents the Euclidean distance between k and v, and S(u,v) represents the Euclidean distance between u and v.

7. The anti-fraud method based on elderly behavioral characteristics as claimed in claim 1, characterized in that: The marked frequencies of all the numbers are measured by the suspiciousness, and the calculation formula of the suspiciousness is: In the formula, V represents the suspiciousness, L represents the number of times the number is marked, and N represents the number of recent calls made by the number.

8. An anti-fraud system based on the behavioral characteristics of the elderly, characterized in that: include: Network building module, used to build a twin network based on the LSTM module; The network training module is used to train the twin network using the historical communication voice data of the elderly to obtain a trained twin network; A relationship graph construction module is used to calculate the similarity between nodes using call records as nodes, and use the similarity as the edge between nodes to construct a call behavior relationship graph; The similarity between the computing nodes is specifically: Extract the behavior feature vector of the call object corresponding to the call record; use the trained twin network to extract the voice feature vector of the call voice corresponding to the call record; Concatenate the behavior feature vector of each node with the corresponding speech feature vector, and obtain a concatenated feature vector for each node; The similarity between the concatenated feature vectors is used as the similarity between the nodes; The number marking module is used to perform cluster analysis on the nodes in the call behavior relationship graph based on the similarity between the nodes to obtain the outlier degree of the node. If the outlier degree is higher than the set threshold, the number corresponding to the node is marked once; The reminder module is used to obtain the marked frequency of all numbers in the recent database. If the marked frequency is higher than the set threshold, it is considered to be a suspicious number and the elderly are reminded.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the anti-fraud method based on the behavioral characteristics of the elderly as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: The computer program enables the computer to execute the anti-fraud method based on the behavioral characteristics of the elderly as described in any one of claims 1-7.

Images