User transaction risk identification method, device and server

Abstract

This specification provides a method, device and server for identifying transaction risks of users. Based on this method, facial video data of the target user to be detected can be first obtained, and a target picture group containing multiple facial pictures arranged in chronological order can be extracted based on the facial video data; then, a preset emotion recognition model can be called to process the target picture group, and the emotional state parameters of the target user can be calculated based on the emotional state changes represented by the multiple facial pictures arranged in chronological order; and then the preset transaction risk identification model can be called to determine whether the transaction behavior in which the target user participates has a preset transaction risk based on the above-mentioned emotional state parameters. In this way, users whose current transaction behaviors have preset transaction risks can be identified more accurately and efficiently, so that corresponding reminders can be given to such users in a timely manner to protect the transaction security of users.

Description

Technical Field

[0001] This specification belongs to the field of artificial intelligence technology, and in particular, relates to a method, device and server for identifying user transaction risks. Background Art

[0002] In the banking business processing scenario, it is usually necessary to rely on bank staff to manually observe the facial expressions of users who come to handle business based on their personal experience and knowledge reserves to determine whether the user is at risk of being defrauded when handling transactions such as transfers and remittances. If it is determined that the user has the above transaction risks, the staff will promptly remind and dissuade the user to avoid the user's property from suffering losses.

[0003] However, the above processes mostly rely on the personal experience and knowledge reserves of the staff, and are easily interfered and influenced by the staff's personal subjective factors, resulting in low reliability and often unable to accurately and efficiently identify users with transaction risks.

[0004] To address the above problems, no effective solution has been proposed yet. Summary of the invention

[0005] This specification provides a user transaction risk identification method, device and server, which can more accurately and efficiently identify users whose transaction behaviors have preset transaction risks, so as to timely remind the users accordingly and protect the users' transaction security.

[0006] This specification provides a method for identifying transaction risks of users, including:

[0007] Obtain facial video data of the target user;

[0008] Extracting a target picture group according to the facial video data; wherein the target picture group includes a plurality of facial pictures arranged in chronological order;

[0009] Calling a preset emotion recognition model to process the target picture group to obtain an emotion state parameter of a target user; wherein the preset emotion recognition model is used to determine the emotion state parameter of the target user based on the emotion state changes represented by a plurality of facial pictures arranged in chronological order;

[0010] Calling a preset transaction risk identification model to process the emotional state parameters of the target user to obtain a corresponding target processing result;

[0011] According to the target processing result, it is determined whether the transaction behavior in which the target user participates has a preset transaction risk.

[0012] In one embodiment, the preset emotion recognition model at least includes: a face facial action unit recognition module and an emotion state parameter calculation module, wherein the face facial action unit recognition module is connected to the emotion state parameter calculation module; wherein the face facial action unit recognition module is used to perform recognition processing on each face picture in a plurality of face pictures included in a target picture group, so as to determine the action features of each face facial action unit in each face picture; and determine the emotion features of each face picture according to the action features of each face facial action unit in each face picture; and the emotion state parameter calculation module is used to calculate the emotion state parameters of the target user according to preset processing rules and the emotion features of each face picture.

[0013] In one embodiment, the facial action unit of the human face includes at least one of the following: eyes, mouth, and eyebrows.

[0014] In one embodiment, the emotional state parameter calculation module is specifically used to arrange the emotional features of multiple facial images in chronological order according to preset processing rules to obtain corresponding emotional feature columns; based on time parameters, the emotional feature columns are integrated to obtain the emotional state parameters of the target user.

[0015] In one embodiment, the preset emotion recognition model also includes: a face positioning module; wherein the face positioning module is used to locate and mark a target image area containing the face of the target user in a facial image; correspondingly, the face facial action unit recognition module is used to perform recognition processing on the target image area.

[0016] In one embodiment, the preset transaction risk includes: risk of fraud.

[0017] In one embodiment, after determining, based on the target processing result, that the transaction behavior in which the target user participates has a preset transaction risk, the method further includes:

[0018] Display relevant risk warning information to the target user; and obtain descriptive data about the transaction behavior provided by the target user;

[0019] According to the description data, it is determined whether to interrupt the transaction behavior in which the target user participates.

[0020] In one embodiment, when it is determined according to the target processing result that the transaction behavior in which the target user participates does not have a preset transaction risk, the method further includes:

[0021] The facial video data of the target user is re-acquired at preset time intervals.

[0022] In one embodiment, the preset emotion recognition model is obtained in the following manner:

[0023] Acquire a sample picture group; wherein the sample picture group includes facial pictures of a plurality of sample faces arranged in chronological order;

[0024] According to a preset labeling rule, the facial images of the sample faces in the sample image group are labeled accordingly to obtain a labeled sample image group;

[0025] The labeled sample image group is used to perform model training to obtain the preset emotion recognition model.

[0026] In one embodiment, according to a preset labeling rule, the facial images of the sample faces in the sample image group are labeled accordingly to obtain the labeled sample image group, including:

[0027] According to a preset labeling rule, determining and labeling facial action units and action features of facial action units in the facial image of the sample face;

[0028] According to the preset mapping relationship, based on the action features of the facial action unit, the emotional feature labels of the facial action unit, the emotional state labels of the facial images of the sample faces, and the emotional state labels of the sample image groups are annotated to obtain the annotated sample image groups.

[0029] In one embodiment, the preset transaction risk identification model is obtained in the following manner:

[0030] Determine risk labels for sample image groups;

[0031] Calling a preset emotion recognition model to process the sample picture group to obtain an emotion state parameter of the sample picture group;

[0032] Combining the risk label and the emotional state parameter of the same sample picture group to obtain a risk training data group corresponding to the sample picture group;

[0033] The risk training data set is used to perform model training to obtain a preset transaction risk identification model.

[0034] This specification also provides a user transaction risk identification device, including:

[0035] An acquisition module, used to acquire facial video data of a target user;

[0036] An extraction module, used for extracting a target picture group according to the facial video data; wherein the target picture group includes a plurality of facial pictures arranged in chronological order;

[0037] A first calling module is used to call a preset emotion recognition model to process the target picture group to obtain the emotion state parameters of the target user; wherein the preset emotion recognition model is used to determine and determine the emotion state parameters of the target user based on the emotion state changes represented by a plurality of facial pictures arranged in chronological order;

[0038] A second calling module is used to call a preset transaction risk identification model to process the emotional state parameters of the target user to obtain a corresponding target processing result;

[0039] The determination module is used to determine whether the transaction behavior in which the target user participates has a preset transaction risk according to the target processing result.

[0040] The present specification also provides a server, comprising a processor and a memory for storing processor executable instructions, wherein the processor implements relevant steps of a method for identifying transaction risks of a user when executing the instructions.

[0041] The present specification also provides a computer-readable storage medium having computer instructions stored thereon, which, when executed, implement the relevant steps of the method for identifying transaction risks of a user.

[0042] The present specification provides a method, device and server for identifying transaction risks of a user. Before implementation, a preset emotion recognition model can be pre-trained to calculate the user's emotional state parameters based on the emotional changes represented by multiple facial pictures arranged in chronological order; at the same time, a preset transaction risk recognition model can be pre-trained to predict whether the transaction behavior in which the user participates has transaction risks based on the user's emotional state parameters; during implementation, the facial video data of the target user to be detected can be first obtained, and a target picture group containing multiple facial pictures arranged in chronological order can be extracted based on the facial video data; then, the preset emotion recognition model can be called to process the target picture group to calculate the emotional state parameters of the target user; and then the preset transaction risk recognition model can be called to determine whether the transaction behavior in which the target user participates has preset transaction risks based on the above-mentioned emotional state parameters. In this way, users whose current transaction behaviors have preset transaction risks can be identified and found more accurately and efficiently, so that the users can be reminded in time to protect the transaction security of the users. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of this specification, the drawings required for use in the embodiments will be briefly introduced below. The drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0044] Figure 1 It is a schematic diagram of an embodiment of the system structure composition of the user transaction risk identification method provided by the embodiment of this specification;

[0045] Figure 2 It is a flowchart of a method for identifying transaction risks of a user provided by an embodiment of this specification;

[0046] Figure 3 It is a schematic diagram of the structure of a server provided by an embodiment of this specification;

[0047] Figure 4 It is a schematic diagram of the structure of a user transaction risk identification device provided by an embodiment of this specification;

[0048] Figure 5 It is a schematic diagram of an embodiment of a method for identifying transaction risks of a user provided by an embodiment of this specification, in a scenario example;

[0049] Figure 6 It is a schematic diagram of an embodiment of a method for identifying transaction risks of a user provided by an embodiment of this specification, in a scenario example. DETAILED DESCRIPTION

[0050] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0051] The present specification provides a method for identifying transaction risks of a user, which can be applied to a system including a server and a terminal device. Figure 1 As shown, the terminal device and the server can be connected via wired or wireless means to perform specific data interaction.

[0052] In this embodiment, the server may specifically include a background server applied to a data processing platform side that can realize functions such as data transmission and data processing. Specifically, the server may be, for example, an electronic device with data calculation, storage and network interaction functions. Alternatively, the server may also be a software program running in the electronic device to provide support for data processing, storage and network interaction. In this embodiment, the number of servers included in the server is not specifically limited. The server may specifically be one server, or several servers, or a server cluster formed by several servers.

[0053] In this embodiment, the terminal device may specifically include a front-end device that is arranged in a transaction service area (e.g., a bank's business hall, etc.), has a built-in or external camera, and can realize functions such as video (or image) data acquisition and data transmission. Specifically, the terminal device may be, for example, a monitor arranged in a bank's business hall, or a smart phone or tablet computer equipped with a camera used by a staff member responsible for handling relevant services for a user, or an automatic service processing device equipped with a camera (e.g., an ATM machine), etc.

[0054] In this embodiment, when the target user goes to the transaction service processing area and uses an ATM machine to process related transaction services (for example, transfer services), the terminal device can regularly collect the facial video data of the target user when handling the business at preset time periods (for example, 3 seconds); and send the facial video data to the server via wired or wireless means.

[0055] Correspondingly, the server receives and obtains the facial video data of the target user.

[0056] First, the server can perform frame-decoding on the facial video data to extract a corresponding target picture group (or picture stream). The target picture group includes a plurality of facial pictures arranged in chronological order, and each facial picture specifically includes facial images of the target user at different time points.

[0057] Next, the server can input the target picture group into a preset emotion recognition model, and run the preset emotion recognition model to output the target user's emotion state parameters. The preset emotion recognition model at least includes a model structure such as a facial action unit recognition module and an emotion state parameter calculation module. Moreover, based on the preset emotion recognition model, the target user's emotion state parameters can be calculated based on the input target picture group and the emotion state changes represented by multiple facial pictures arranged in chronological order.

[0058] Furthermore, the server may input the emotional state parameters of the target user into a preset transaction risk identification model, and run the preset transaction risk identification model to output a probability value as to whether the transaction behavior in which the target user is currently involved has a preset transaction risk, as a target processing result.

[0059] Furthermore, the server can determine whether the transaction behavior currently being participated in by the target user has a preset transaction risk according to the above target processing result. For example, it can be determined that the target user who is currently handling a transfer business is suspected of being defrauded.

[0060] When it is determined that the current target user does not have the preset transaction risk, the server can continue to obtain the facial video data of the target user for the next preset time period, and repeat the above process to continue to identify and judge whether the transaction behavior in which the target user participates has the preset transaction risk.

[0061] If it is determined that the current target user has a preset transaction risk, a prompt message can be sent to the staff in the transaction business handling area so that the staff can remind the target user of the relevant transaction risks based on the above prompt message. At the same time, the staff can also manually inquire the target user to obtain more information to further and more accurately determine whether the target user really has relevant transaction risks. If it is determined that the target user really has (or has a high probability of having) relevant transaction risks, the staff can promptly dissuade the target user, thereby protecting the target user's transaction property safety and preventing the target user from suffering losses.

[0062] Based on the above system, the server can more accurately and efficiently identify users whose transaction behaviors involve preset transaction risks by combining the preset emotion recognition model and the preset transaction risk identification model, and can then promptly provide corresponding reminders to the user to protect the user's transaction security.

[0063] See also Figure 2 As shown, the embodiment of this specification provides a method for identifying transaction risks of a user. The method is specifically applied to the server side. When implemented specifically, the method may include the following contents.

[0064] S201: Acquire facial video data of a target user;

[0065] S202: extracting a target picture group according to the facial video data; wherein the target picture group includes a plurality of facial pictures arranged in chronological order;

[0066] S203: calling a preset emotion recognition model to process the target picture group to obtain an emotion state parameter of the target user; wherein the preset emotion recognition model is used to determine the emotion state parameter of the target user based on the emotion state changes represented by the multiple facial pictures arranged in chronological order;

[0067] S204: calling a preset transaction risk identification model to process the emotional state parameters of the target user to obtain a corresponding target processing result;

[0068] S205: Determine, based on the target processing result, whether the transaction behavior in which the target user participates has a preset transaction risk.

[0069] Through the above embodiment, the server can first use the preset emotion recognition model based on the emotion state changes represented by multiple facial images arranged in chronological order to accurately calculate the emotion state parameters of the target user; and then use the preset transaction risk recognition model to accurately determine whether the transaction behavior involved by the target user has a preset transaction risk based on the above emotion state parameters. In this way, users whose transaction behaviors involve preset transaction risks can be identified more accurately and efficiently, so that corresponding reminders can be given to such users in a timely manner.

[0070] In some embodiments, the target user may specifically be a user who is currently participating in a transaction, for example, a user who is currently handling a transaction such as transfer, remittance, etc. at a bank's business hall.

[0071] In some embodiments, the terminal device may collect the target user's facial video data within a preset time period (e.g., 5 seconds) at intervals, and send the target user's facial video data to the server via the network. Correspondingly, the server may receive and obtain the target user's facial video data.

[0072] During specific implementation, the server may also intercept a segment of time-frequency data including the face image of the target user from a whole segment of video data provided by the terminal device as the facial video data of the target user.

[0073] In some embodiments, when the server is implemented, the target user's facial video data can be deframed to extract the target picture group. The target picture group may specifically include multiple facial pictures arranged in chronological order (for example, the order of the time points corresponding to each facial picture). Each of the multiple facial pictures corresponds to a time point when the pictures are collected, and each facial picture may include a facial image of the target user collected at the corresponding time point.

[0074] In some embodiments, the above-mentioned preset emotion recognition model can be specifically understood as a pre-trained one, which is used to determine and output emotional state parameters for describing the user's emotional state based on corresponding emotional changes according to multiple facial images input into the model.

[0075] In some embodiments, the above-mentioned preset transaction risk identification model can be specifically understood as a pre-trained model that is used to output a probability value representing the preset transaction risk that the user may have when performing the current transaction behavior based on the emotional state parameters of the input model.

[0076] In some embodiments, the preset emotion recognition model at least includes: a face action unit recognition module, an emotion state parameter calculation module and other model structures, and the face action unit recognition module is connected to the emotion state parameter calculation module. The face action unit recognition module can be used to perform recognition processing on each of the multiple face pictures included in the target picture group to determine the action features of each face action unit in each face picture; and determine the emotion features of each face picture based on the action features of each face action unit in each face picture. The emotion state parameter calculation module can be used to calculate the emotion state parameters of the target user based on the preset processing rules and the emotion features of each face picture.

[0077] Through the above embodiments, the preset emotion recognition model can be used in combination with the facial action unit recognition module and the emotion state parameter calculation module. First, the action features of each facial action unit in each facial image are recognized and extracted to obtain the emotion features for a single facial image. Then, according to the preset processing rules, the emotion features of multiple facial images are comprehensively utilized to calculate the emotion state parameters with higher reference value and better effect.

[0078] In some embodiments, the above-mentioned facial action recognition module can be specifically understood as a pre-trained convolutional neural network structure based on micro-expression technology.

[0079] When the preset emotion recognition model is specifically run, the facial action unit recognition module will process each of the multiple facial images separately.

[0080] Taking the facial action unit recognition module processing the current facial picture among multiple facial pictures as an example, first, each facial action unit in the current facial picture can be determined; then, the action features of each facial action unit in the current facial picture can be extracted; then, based on the preset mapping relationship, the emotional features of each facial action unit can be determined according to the action features of each facial action unit; then, the emotional features of multiple facial action units in the current facial picture can be combined, and the emotional state parameters of the current facial picture can be calculated and output through weighted operations.

[0081] In some embodiments, the face action unit may include at least one of the following: eyes, mouth, eyebrows, etc. Of course, the face action units listed above are only a schematic illustration. In specific implementation, other face action units such as cheeks, forehead, etc. may also be introduced according to specific circumstances and processing requirements.

[0082] The above-mentioned facial action unit can be specifically understood as a position area on the face where the action generated is highly correlated with the user's emotional changes based on the micro-expression theory.

[0083] Through the above embodiments, the action features of some facial action units of a person's face can be collected in a targeted manner to more accurately and efficiently determine the emotional state parameters of the user.

[0084] In some embodiments, the collected motion features may be different for different facial motion units. Specifically, for eyes, the collected motion features may include: closed, half-open, fully open, etc. For eyebrows, the collected motion features may include: frowning, stretching, etc. For mouths, the collected motion features may include: mouth corners raised, mouth corners pulled down, etc.

[0085] In specific implementation, the above action features can be encoded according to the preset action feature encoding rules to obtain data values ​​corresponding to the above action features. For example, the data value corresponding to the corner of the mouth is 1001, the data value corresponding to the frown is 2002, and so on. In this way, the above data values ​​can be used to replace the action features in the future to carry out specific model training and model calculation.

[0086] In some embodiments, the above-mentioned emotional characteristics may specifically include: anger, sadness, calmness, happiness, etc.

[0087] In specific implementation, the above emotional features can also be encoded according to the preset emotional encoding rules to obtain data values ​​corresponding to the above emotional features. For example, the data value corresponding to anger is 0, the data value corresponding to sadness is 0.3, the data value corresponding to calmness is 0.5, the data value corresponding to happiness is 0.7, etc. In this way, the above data values ​​can be used to replace emotional features in the future to perform specific model training and model calculation.

[0088] It should be added that the data value obtained by encoding the emotion feature based on the above preset emotion encoding rules is a value greater than or equal to 0, which can effectively reduce the error in the subsequent training model during operation.

[0089] In some embodiments, it should be supplemented that, based on a preset mapping relationship, for different facial action units, the mapping rules between action features and emotion features may be different.

[0090] When performing weighted operations to calculate the emotional state parameters of a facial image, the weight values ​​corresponding to the facial action units of different faces may be set differently. For example, since the emotional characteristics of eyebrows can more clearly reflect the emotional state of the user than the emotional characteristics of the mouth, the weight value of eyebrows can be set to a value greater than the weight value of the mouth during weighted operations. In this way, relatively accurate emotional state parameters for the facial image can be obtained.

[0091] Furthermore, corresponding weight values ​​can be configured for each facial action unit according to the emotional characteristics of different facial action units. For example, based on anatomy, it is considered that the muscles of the mouth are more important than the muscles of the eyes in expressing emotions such as joy. Therefore, when the emotional characteristics of the mouth and the emotional characteristics of the eyes in a certain facial image are close to being happy, the weight value of the mouth can be set larger than the weight of the eyes during weighted calculation. In this way, a relatively more accurate emotional state parameter for the facial image can be obtained.

[0092] In some embodiments, the above-mentioned preset processing rule may specifically be a calculation rule based on integration.

[0093] In some embodiments, the emotional state parameter calculation module can be specifically used to arrange the emotional features of multiple facial images in chronological order according to preset processing rules to obtain corresponding emotional feature columns; and then integrate the emotional feature columns based on time parameters to obtain the emotional state parameters of the target user.

[0094] Based on the above embodiment, according to preset processing rules, by performing time point-based integration processing on the emotional features of multiple facial images arranged in chronological order along the time axis, emotional state parameters that can effectively reflect the emotional changes of the user during the time period can be obtained.

[0095] In some embodiments, the above-mentioned integration processing of the emotion feature column based on the time parameter to obtain the emotion state parameter of the target user may include the following contents during specific implementation:

[0096] X=∫xidt

[0097] Wherein, X is the emotional state parameter of the target user, xi is the emotional feature of the face picture numbered i among multiple face pictures, and t is time.

[0098] In some embodiments, the above-mentioned preset emotion recognition model may further include: a face positioning module; wherein the face positioning module is used to locate and mark the target image area containing the face of the target user in the facial picture; correspondingly, the face facial action unit recognition module is used to identify and process the target image area.

[0099] Through the above embodiments, when the preset emotion recognition model is specifically run, the face positioning module can first find a relatively small target image area that only contains the face of the target user in the face picture; and then the face action unit recognition module can perform targeted recognition processing on the above target image area, without the need to perform recognition processing on all image areas of the face picture, thereby reducing the data processing amount of the model to a certain extent and improving the overall data processing efficiency.

[0100] In some embodiments, the preset transaction risks may specifically include: risk of fraud, etc. Of course, in specific implementation, according to specific application scenarios and processing requirements, the preset transaction risks may also include other types of risks. For example, the target user is currently involved in a transaction involving the risk of suspected illegal activities such as money laundering or gambling. For another example, the target user is currently involved in a transaction involving the risk of personal privacy being leaked, etc.

[0101] Through the above embodiments, the user transaction risk identification method provided in this specification can be expanded and applied to various application scenarios to perform real-time detection and identification of various different types of transaction risks.

[0102] In some embodiments, when the preset transaction risk includes the risk of fraud, after determining, based on the target processing result, that the transaction behavior in which the target user participates has the preset transaction risk, the method, when implemented specifically, may also include the following: displaying relevant risk warning information to the target user; and obtaining descriptive data about the transaction behavior provided by the target user; and determining, based on the descriptive data, whether to interrupt the transaction behavior in which the target user participates.

[0103] Through the above embodiments, when it is detected that the transaction behavior currently participated by the target user is at risk of fraud, the target user can be promptly warned of the risk, thereby effectively preventing the user from being defrauded, protecting the user's transaction security, and preventing the user from suffering property losses.

[0104] In some embodiments, during specific implementation, after determining that there is a preset transaction risk, the server can generate and display a first interactive interface to the user. The first interactive interface displays risk warning information related to the preset transaction risk to remind the target user that the transaction behavior currently involved may be risky. At the same time, the server can also generate and display a second interactive interface to the target user. The second interactive interface includes a description data input box, and near the description data input box, it also displays a guide for the user to use text language to enter text in the above description database to describe the current transaction behavior (including behavior content, behavior reasons, etc.). Accordingly, the server can collect and obtain the descriptive data on the transaction behavior provided by the target user through the second interactive interface.

[0105] Furthermore, the server can call the pre-selected trained semantic recognition model to process the description data and obtain the corresponding semantic recognition result. And according to the semantic recognition result, it is further confirmed whether the transaction behavior currently participated by the target user really has the preset transaction risk. If it is confirmed that the transaction behavior currently participated by the target user really has the preset transaction risk, then the server can dissuade the target user by notifying nearby staff, or interrupt the network connection of the terminal device that the user currently relies on when conducting the transaction, interrupt the transaction behavior currently participated by the target user, so as to protect the user's transaction security.

[0106] In some embodiments, when it is determined based on the target processing result that the transaction behavior in which the target user participates does not have a preset transaction risk, the method, when implemented, may also include: reacquiring facial video data of the target user at preset time intervals.

[0107] In this way, the server can trigger the collection of facial video data of the target user in the next preset time period, and repeat the above detection and identification process based on the facial video data of the target user in the next preset time period, so as to continuously detect and identify whether the transaction behavior in which the target user participates has the preset transaction risk.

[0108] In some embodiments, when it is determined that the transaction behavior in which the target user participates does not have a preset transaction risk according to the target processing result of the current preset time period, the method may further include: obtaining the emotional state parameters of the current preset time period obtained based on the facial video data of the target user in the current preset time period; at the same time, querying the historical records to obtain the emotional state parameters of the previous preset time period; calculating the emotional state change values ​​of two adjacent preset time periods according to the emotional state parameters of the current preset time period and the emotional state parameters of the previous preset time period; and then determining whether the transaction behavior in which the target user participates has a preset transaction risk according to the above emotional state change values. Thus, the preset transaction risk can be identified more accurately.

[0109] In some embodiments, the preset emotion recognition model can be obtained in the following manner:

[0110] S1: Obtain a sample picture group; wherein the sample picture group includes facial pictures of a plurality of sample faces arranged in chronological order;

[0111] S2: According to a preset labeling rule, the facial images of the sample faces in the sample image group are labeled accordingly to obtain a labeled sample image group;

[0112] S3: Performing model training using the labeled sample image group to obtain the preset emotion recognition model.

[0113] Through the above embodiments, a preset emotion recognition model with better effect and higher accuracy can be pre-built and trained.

[0114] In some embodiments, a first initial model based on a neural network may be constructed first; and then the first initial model may be supervised and learned using the labeled sample image group to obtain a preset emotion recognition model.

[0115] In some embodiments, after obtaining the sample picture group, the sample picture group can be divided into two data sets, and one of the data sets can be labeled as a training set; the other data set can be used as a test set; and then the training set and the test set can be used to perform model training to obtain a preset emotion recognition model.

[0116] In some embodiments, the facial images of the sample faces in the sample image group are labeled accordingly according to the preset labeling rules to obtain the labeled sample image group. When implemented specifically, the following contents may be included:

[0117] S1: according to a preset labeling rule, determine and label the facial action unit and the action features of the facial action unit in the facial image of the sample face;

[0118] S2: According to the preset mapping relationship, based on the action features of the facial action unit, the emotional feature label of the facial action unit, the emotional state label of the facial image of the sample face, and the emotional state label of the sample image group are marked to obtain the marked sample image group.

[0119] Through the above embodiments, a labeled sample picture group with relatively good training effect can be obtained according to the preset labeling rules, and then the labeled sample picture group can be used to train a preset emotion recognition model with higher accuracy.

[0120] In some embodiments, the preset transaction risk identification model can be obtained in the following manner:

[0121] S1: Determine the risk label of the sample image group;

[0122] S2: calling a preset emotion recognition model to process the sample picture group to obtain an emotion state parameter of the sample picture group;

[0123] S3: combining the risk label and the emotional state parameter of the same sample picture group to obtain a risk training data group corresponding to the sample picture group;

[0124] S4: Performing model training using the risk training data set to obtain a preset transaction risk identification model.

[0125] Through the above embodiments, a preset transaction risk identification model with better effect and higher accuracy can be pre-built and trained.

[0126] In some embodiments, during specific implementation, a second initial model based on a convolutional neural network can be first constructed. After obtaining the risk training data group, the risk training data group can be divided into two data groups, which are respectively recorded as training combination verification groups. During specific training, the training group can be used to train the second initial model to obtain the trained third model. Then, the verification group can be input into the third model to obtain the corresponding verification result data; the ROC curve can be used to perform a performance test on the verification result data; according to the performance test results, combined with the AUC value as an aid, the model parameters of the third model are adjusted to obtain a preset transaction risk identification model that meets the requirements.

[0127] As can be seen from the above, based on the transaction risk identification method for users provided in the embodiments of this specification, before specific implementation, a preset emotion recognition model can be pre-trained to calculate the user's emotional state parameters based on the emotional changes represented by multiple facial pictures arranged in chronological order; at the same time, a preset transaction risk identification model can be pre-trained to predict whether the transaction behavior in which the user participates has transaction risks based on the user's emotional state parameters; during specific implementation, the facial video data of the target user to be detected can be first obtained, and a target picture group containing multiple facial pictures arranged in chronological order can be extracted based on the facial video data; then, the preset emotion recognition model can be called to process the target picture group to calculate the emotional state parameters of the target user; and then the preset transaction risk identification model can be called to determine whether the transaction behavior in which the target user participates has preset transaction risks based on the above-mentioned emotional state parameters. In this way, users whose transaction behaviors have preset transaction risks can be found more accurately and efficiently, so that the users can be reminded accordingly in a timely manner to protect the user's transaction security.

[0128] The embodiment of the present specification also provides a server, including a processor and a memory for storing processor executable instructions, and the processor can perform the following steps according to the instructions when implemented: acquiring facial video data of a target user; extracting a target picture group based on the facial video data; wherein the target picture group includes a plurality of facial pictures arranged in chronological order; calling a preset emotion recognition model to process the target picture group to obtain emotional state parameters of the target user; wherein the preset emotion recognition model is used to determine and determine the emotional state parameters of the target user based on the emotional state changes represented by the plurality of facial pictures arranged in chronological order; calling a preset transaction risk identification model to process the emotional state parameters of the target user to obtain a corresponding target processing result; and determining whether the transaction behavior in which the target user participates has a preset transaction risk based on the target processing result.

[0129] In order to complete the above instructions more accurately, refer to Figure 3 As shown, the embodiment of this specification also provides another specific server, wherein the server includes a network communication port 301, a processor 302 and a memory 303, and the above structures are connected through internal cables so that each structure can perform specific data interaction.

[0130] The network communication port 301 can be used to obtain facial video data of a target user.

[0131] The processor 302 can be specifically used to extract a target picture group based on the facial video data; wherein the target picture group includes multiple facial pictures arranged in chronological order; call a preset emotion recognition model to process the target picture group to obtain emotional state parameters of a target user; wherein the preset emotion recognition model is used to determine and determine the emotional state parameters of the target user based on the emotional state changes represented by the multiple facial pictures arranged in chronological order; call a preset transaction risk identification model to process the emotional state parameters of the target user to obtain a corresponding target processing result; and determine whether the transaction behavior in which the target user participates has a preset transaction risk based on the target processing result.

[0132] The memory 303 may be specifically used to store corresponding instruction programs.

[0133] In this embodiment, the network communication port 301 can be a virtual port that is bound to different communication protocols so that different data can be sent or received. For example, the network communication port can be a port responsible for web data communication, a port responsible for FTP data communication, or a port responsible for email data communication. In addition, the network communication port can also be a physical communication interface or communication chip. For example, it can be a wireless mobile network communication chip, such as GSM, CDMA, etc.; it can also be a Wifi chip; it can also be a Bluetooth chip.

[0134] In this embodiment, the processor 302 may be implemented in any appropriate manner. For example, the processor may take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (such as software or firmware) executable by the (micro)processor, a logic gate, a switch, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller, etc. This specification does not limit this.

[0135] In this embodiment, the memory 303 may include multiple levels. In a digital system, anything that can store binary data can be a memory; in an integrated circuit, a circuit with a storage function but no physical form is also called a memory, such as RAM, FIFO, etc.; in a system, a storage device with a physical form is also called a memory, such as a memory stick, TF card, etc.

[0136] The embodiments of this specification also provide a computer storage medium based on the transaction risk identification method for the above-mentioned user, wherein the computer storage medium stores computer program instructions, which, when executed, implement the following: acquiring facial video data of a target user; extracting a target picture group based on the facial video data; wherein the target picture group includes a plurality of facial pictures arranged in chronological order; calling a preset emotion recognition model to process the target picture group to obtain emotional state parameters of the target user; wherein the preset emotion recognition model is used to determine and determine the emotional state parameters of the target user based on the emotional state changes represented by the plurality of facial pictures arranged in chronological order; calling a preset transaction risk identification model to process the emotional state parameters of the target user to obtain a corresponding target processing result; and determining, based on the target processing result, whether the transaction behavior in which the target user participates has a preset transaction risk.

[0137] In this embodiment, the storage medium includes, but is not limited to, a random access memory (RAM), a read-only memory (ROM), a cache, a hard disk (HDD), or a memory card. The memory may be used to store computer program instructions. The network communication unit may be an interface for network connection communication set in accordance with the standard specified by the communication protocol.

[0138] In this embodiment, the functions and effects specifically implemented by the program instructions stored in the computer storage medium can be explained in comparison with other implementations and will not be described in detail here.

[0139] See also Figure 4 As shown, at the software level, the embodiment of this specification also provides a user transaction risk identification device, which may specifically include the following structural modules:

[0140] The acquisition module 401 may be specifically used to acquire facial video data of a target user;

[0141] The extraction module 402 may be specifically configured to extract a target picture group according to the facial video data; wherein the target picture group includes a plurality of facial pictures arranged in chronological order;

[0142] The first calling module 403 may be specifically used to call a preset emotion recognition model to process the target picture group to obtain the emotion state parameter of the target user; wherein the preset emotion recognition model is used to determine and determine the emotion state parameter of the target user based on the emotion state changes represented by the multiple facial pictures arranged in chronological order;

[0143] The second calling module 404 may be specifically used to call a preset transaction risk identification model to process the emotional state parameters of the target user to obtain a corresponding target processing result;

[0144] The determination module 405 may be specifically used to determine whether the transaction behavior in which the target user participates has a preset transaction risk according to the target processing result.

[0145] It should be noted that the units, devices or modules described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. For the convenience of description, the above devices are described separately by functions divided into various modules. Of course, when implementing this specification, the functions of each module can be implemented in the same or more software and / or hardware, or the modules that implement the same function can be implemented by a combination of multiple sub-modules or sub-units. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0146] As can be seen from the above, the user transaction risk identification device provided in the embodiments of this specification can more accurately and efficiently discover users whose transaction behaviors have preset transaction risks, so as to promptly remind the users accordingly to protect the users' transaction security.

[0147] In a specific scenario example, the transaction risk identification method provided in this specification can be used to build an anti-fraud counter system based on emotion recognition to detect online whether a customer handling business at a bank counter is suspected of being defrauded. For the specific implementation process, please refer to the following content.

[0148] For details, please refer to Figure 5 As shown, the counter uses a camera to obtain a real-time facial video of the customer (e.g., facial video data), and sends the video to the server at a certain time interval (e.g., a preset time period). For example, it can be set to send the recorded video to the server every 3 seconds for subsequent emotion analysis.

[0149] After receiving the sent video, the server deframes the video and then inputs each frame of the image into the customer fraud risk index model (equivalent to a combination of a preset emotion recognition model and a preset transaction risk recognition model) to perform emotion recognition first; then, based on the emotion recognition results, the customer fraud risk index (for example, the probability value of the presence of a preset transaction risk) is calculated.

[0150] The teller interface can display the calculated customer fraud risk index in real time. If the index exceeds a certain threshold (for example, 70%), the teller is required to suspend the transaction and communicate with the customer to understand the reason for the customer's transaction and whether there is a historical transaction initiated for similar reasons.

[0151] If the customer's transaction is confirmed to be normal, the teller can assist the customer to complete the transaction. Otherwise, the transaction needs to be terminated and the transaction situation needs to be recorded.

[0152] The customer fraud risk index model used includes: a video decompression processing module, an emotion dynamic recognition module (eg, a preset emotion recognition model), and a customer fraud risk index calculation module (eg, a preset transaction risk recognition model).

[0153] Run the model, the model can deframe the video and obtain the photo stream corresponding to the video (for example, a target image group containing multiple facial images arranged in sequence). Then use the face recognition and positioning model in the emotional dynamics recognition module to locate the face position in the image. Then, the facial action unit recognition model in the emotional dynamics recognition module can be used to identify the facial unit action sequence of the face and obtain the emotional state value of the face. Finally, the integral formula in the emotional dynamics recognition module is used to calculate the emotional state value of the video (for example, emotional state parameters).

[0154] The face recognition and positioning model may specifically use OpenCV (a packaged computer vision function library) to lock the face area of ​​each frame of the acquired facial expression video, and may specifically include operations such as face capture, calibration, and image scaling.

[0155] The facial action unit recognition module can specifically extract the features of each picture in the picture sequence through a pre-trained convolutional neural network based on micro-expression recognition technology to obtain a series of feature sequences; then these feature sequences are input into a recurrent neural network, and the feature changes in the picture sequence can be analyzed to identify the facial action unit combination of the face at each moment. By using the facial action unit recognition model trained by micro-expression recognition technology, continuous facial expressions can be detected and the emotional changes therein can be identified (the emotional features of the face picture can be obtained).

[0156] The training of the above-mentioned emotion dynamic recognition module can be specifically divided into two stages: data preparation and model training.

[0157] Among them, the above data preparation includes: according to the principles of human anatomy, dividing the human face into different facial action units, and using the corresponding facial action unit combinations as labels for the image, and inputting the face image and the corresponding facial action unit combinations as data into the model for training.

[0158] The training model includes: dividing the obtained labeled facial action unit data set into a training set and a verification set; then inputting the facial images of the training set and the corresponding facial action unit set into the initial model, training the data of the initial model, and continuously adjusting the parameters to obtain a recognition model with relatively high accuracy and efficiency; then inputting the data of the verification set into the trained recognition model, comparing and verifying the recognized facial unit set with the labeled data of the verification set, if the accuracy of the recognition model meets the requirements, then the recognition model can be used as a facial action unit recognition model, if the accuracy does not meet the requirements, then continue to adjust the parameter training.

[0159] The calculation of the emotional state value of the face is to set different emotional values ​​according to different facial action unit states. According to experiments, different parts of the face have different expression functions. Among them, the eyes are most important for expressing sadness, the mouth is most important for expressing happiness and disgust, and the forehead can provide a signal of surprise. The eyes, mouth and forehead are very important for expressing anger. In addition, experiments have shown that the mouth muscles are more important than the eye muscles for expressing emotions such as joy and resentment, while the eye muscles are more important than the mouth muscles for expressing emotions such as sadness and horror. When people express a real smile, their cheeks rise and the muscles around the eyes are piled up; when people fake a smile, only the muscles of the lips are active and the jaw droops.

[0160] The facial unit action data is trained to obtain the specific emotion values ​​of different faces. After obtaining the specific emotions, the weighted values ​​are calculated according to certain rules. The calculated fraud risk value is compared with the label corresponding to the facial emotion, and the parameters are continuously adjusted to obtain an emotion recognition model with a relatively high accuracy.

[0161] The expression states (action features) and label values ​​(emotion features) of the facial units can be specifically found in Table 1.

[0162] Table 1

[0163]

[0164] When specifically calculating the emotional state value of the video, the facial emotional value of each frame of the video may be integrated using the following formula to calculate the emotional state value calculation module of the video:

[0165] X=∫xidt

[0166] Wherein, X is the emotional state of the customer (for example, the emotional state value of the customer), xi is the emotion value (emotional feature) to be calculated corresponding to the i-th frame image, and t is time.

[0167] The emotional state of the customer is input into the customer fraud risk index calculation module to obtain the possibility (0-100%) of the customer being suspected of being defrauded. The specific process of training the customer fraud risk index calculation module may include: data preparation, model training, and model verification.

[0168] Data preparation: Use historically saved customer videos as the data source for training, and use the customer facial emotion state values ​​corresponding to the videos and the customer fraud label as labeled training data, as shown in Table 2. Divide the data set into training set, validation set, and test set.

[0169] Table 2

[0170] Whether it was defrauded Label yes 1 no 0

[0171] Model training: This model is trained using a convolutional neural network model. The customer's facial emotional state value is used as input to the model for training. The supervised machine learning method is used for training. This model is trained using a convolutional neural network model. During this process, the parameters are continuously adjusted to obtain an emotion recognition model with relatively high accuracy.

[0172] Model validation: Input the validation set data into the trained emotion recognition model, and use the ROC curve to perform performance testing on the obtained results. Since the ROC curve itself is not easy to quantify the model effect, the AUC value is added to assist in quantification. The AUC value represents the area under the ROC curve, which is generally between 0.5 and 1, with 0.5 representing random classification and 1 representing complete classification. When the AUC value is close to 1, the model effect is the best. That is, when the area under the ROC curve is closest to 1, the model effect is optimal.

[0173] Based on the above ideas, the anti-fraud counter system based on emotion recognition was constructed and applied. Figure 6 As shown, it includes: counter camera, server, and counter system.

[0174] The counter camera is a module for implementing video recording. It uses a camera at the counter window to record the video of the customer receiving service from the teller and transmits the video to the server. The server includes two emotion recognition modules and a customer fraud risk index calculation module. When the customer's fraud risk index exceeds the set threshold, the counter system pops up a prompt box on the bank counter system interface, requiring the teller to understand the customer's operation purpose and provide the customer's historical operation records to determine whether the customer is a suspected fraud customer.

[0175] Through the above scenario examples, the anti-fraud counter system based on emotion recognition can automatically detect abnormal behaviors of customers in the process of receiving services, monitor changes in customers' facial expressions in real time, and dynamically monitor changes in customers' emotions. Intelligent recognition can be used instead of manual recognition to effectively identify suspected deceived customers and effectively reduce the occurrence of fraud incidents. Specifically, by obtaining the customer's fraud risk index based on emotion recognition and unifying the standards for identifying suspected defrauded customers, the accuracy of bank staff in identifying deceived customers can be improved, the economic losses of customers can be reduced, the service quality of the bank can be improved, and the trust of customers can be obtained. Through the anti-fraud counter system based on emotion recognition, the abnormal behavior of customers can be monitored in real time without missing the abnormal behavior of customers. The dynamic monitoring of customers' emotions over a period of time can monitor customers' emotions more comprehensively and scientifically. Based on artificial intelligence technology, the human resource training cost of the bank will be reduced. Even staff who have just started working can communicate effectively with customers according to prompts and improve the accuracy of identifying deceived customers.

[0176] Although the present specification provides method operation steps as described in the embodiments or flow charts, more or less operation steps may be included based on conventional or non-creative means. The order of steps listed in the embodiments is only one way of executing the order of many steps, and does not represent a unique execution order. When the device or client product in practice is executed, it can be executed in sequence or in parallel according to the method shown in the embodiments or the drawings (for example, a parallel processor or a multi-threaded processing environment, or even a distributed data processing environment). The term "include", "include" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, product or device including a series of elements includes not only those elements, but also includes other elements that are not explicitly listed, or also includes elements inherent to such a process, method, product or device. In the absence of more restrictions, it is not excluded that there are other identical or equivalent elements in the process, method, product or device including the elements. The first, second, etc. words are used to represent the name, and do not represent any particular order.

[0177] Those skilled in the art also know that, in addition to implementing the controller in a purely computer-readable program code, the controller can be made to implement the same function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, such a controller can be considered as a hardware component, and the devices for implementing various functions included therein can also be considered as structures within the hardware component. Or even, the devices for implementing various functions can be considered as both software modules for implementing the method and structures within the hardware component.

[0178] This specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, classes, etc. that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media including storage devices.

[0179] Through the description of the above embodiments, it can be known that those skilled in the art can clearly understand that the present specification can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present specification can essentially be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a mobile terminal, a server, or a network device, etc.) to execute the methods described in each embodiment of the present specification or some parts of the embodiments.

[0180] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. This specification can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, multi-processor systems, microprocessor-based systems, set-top boxes, programmable electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc.

[0181] Although the present specification is described through embodiments, those skilled in the art will appreciate that there are many modifications and changes to the present specification without departing from the spirit of the present specification, and it is intended that the appended claims include these modifications and changes without departing from the spirit of the present specification.

Claims

1. A method for identifying transaction risks of a user, characterized in that: include: Obtain facial video data of the target user; Extracting a target picture group according to the facial video data; wherein the target picture group includes a plurality of facial pictures arranged in chronological order; Calling a preset emotion recognition model to process the target picture group to obtain the emotion state parameters of the target user; wherein the preset emotion recognition model is used to determine and determine the emotion state parameters of the target user based on the emotion state changes represented by multiple facial pictures arranged in chronological order; wherein the preset emotion recognition model at least includes: a face facial action unit recognition module and an emotion state parameter calculation module, and the face facial action unit recognition is connected to the emotion state parameter calculation module; wherein the face facial action unit recognition module is used to respectively recognize and process each facial picture in the multiple facial pictures included in the target picture group to determine the action features of each facial action unit in each facial picture; and according to the action features of each facial action unit in each facial picture, determine the emotion features of each facial picture; the emotion state parameter calculation module is used to calculate the emotion state parameters of the target user according to the preset processing rules and the emotion features of each facial picture; Calling a preset transaction risk identification model to process the emotional state parameters of the target user to obtain a corresponding target processing result; According to the target processing result, determining whether the transaction behavior in which the target user participates has a preset transaction risk; the preset transaction risk includes: fraud risk; The method of calling a preset emotion recognition model to process the target picture group includes: using a face action unit recognition module to determine each face action unit in each face picture in the target picture group; and extracting the action features of each face action unit in each face picture; based on a preset mapping relationship, determining the emotion features of each face action unit according to the action features of each face action unit; respectively integrating the emotion features of multiple face action units in each face picture, and obtaining the emotion features of each face picture through weighted operation; using an emotion state parameter calculation module according to a preset processing rule, along the time axis, by performing time point-based integration processing on the emotion features of multiple face pictures arranged in chronological order, to obtain an emotion state parameter that can reflect the emotion changes of the target user within a time period; wherein the preset processing rule is a calculation rule based on integration; Among them, after determining that the transaction behavior in which the target user participates has a preset transaction risk according to the target processing result, the method also includes: displaying relevant risk warning information to the target user; and obtaining descriptive data about the transaction behavior provided by the target user; wherein the descriptive data includes descriptive data about the content of the behavior and the cause of the behavior; calling a pre-selected and trained semantic recognition model to process the descriptive data to obtain a corresponding semantic recognition result; based on the semantic recognition result, confirming whether the transaction behavior currently participated by the target user really has a preset transaction risk; if it is confirmed that the transaction behavior currently participated by the target user really has a preset transaction risk, the target user is dissuaded accordingly by notifying nearby staff, or the target user is interrupted by interrupting the network connection of the terminal device on which the user currently relies when performing the transaction.

2. The method according to claim 1, characterized in that The facial action unit of the human face includes at least one of the following: eyes, mouth, and eyebrows.

3. The method according to claim 1, characterized in that: The emotional state parameter calculation module is specifically used to arrange the emotional features of multiple facial images in chronological order according to preset processing rules to obtain corresponding emotional feature columns; based on time parameters, the emotional feature columns are integrated to obtain the emotional state parameters of the target user.

4. The method according to claim 1, characterized in that: The preset emotion recognition model also includes: a face positioning module; wherein the face positioning module is used to locate and mark a target image area containing the face of the target user in a facial image; correspondingly, the face facial action unit recognition module is used to identify and process the target image area.

5. The method according to claim 1, characterized in that When it is determined according to the target processing result that the transaction behavior in which the target user participates does not have a preset transaction risk, the method further includes: The facial video data of the target user is re-acquired at preset time intervals.

6. The method according to claim 1, characterized in that The preset emotion recognition model is obtained in the following manner: Acquire a sample picture group; wherein the sample picture group includes facial pictures of a plurality of sample faces arranged in chronological order; According to a preset labeling rule, the facial images of the sample faces in the sample image group are labeled accordingly to obtain a labeled sample image group; The labeled sample image group is used to perform model training to obtain the preset emotion recognition model.

7. The method according to claim 6, characterized in that According to a preset labeling rule, the facial images of the sample faces in the sample image group are labeled accordingly to obtain a labeled sample image group, including: According to a preset labeling rule, determining and labeling facial action units and action features of facial action units in the facial image of the sample face; According to the preset mapping relationship, based on the action features of the facial action unit, the emotional feature labels of the facial action unit, the emotional state labels of the facial images of the sample faces, and the emotional state labels of the sample image groups are annotated to obtain the annotated sample image groups.

8. The method according to claim 6, characterized in that The preset transaction risk identification model is obtained in the following manner: Determine risk labels for sample image groups; Calling a preset emotion recognition model to process the sample picture group to obtain an emotion state parameter of the sample picture group; Combining the risk label and the emotional state parameter of the same sample picture group to obtain a risk training data group corresponding to the sample picture group; The risk training data set is used to perform model training to obtain a preset transaction risk identification model.

9. A user transaction risk identification device, characterized in that: include: An acquisition module, used to acquire facial video data of a target user; An extraction module, used for extracting a target picture group according to the facial video data; wherein the target picture group includes a plurality of facial pictures arranged in chronological order; A first calling module is used to call a preset emotion recognition model to process the target picture group to obtain the emotion state parameters of the target user; wherein the preset emotion recognition model is used to determine and determine the emotion state parameters of the target user based on the emotion state changes represented by multiple facial pictures arranged in chronological order; wherein the preset emotion recognition model at least includes: a face facial action unit recognition module and an emotion state parameter calculation module, and the face facial action unit recognition is connected to the emotion state parameter calculation module; wherein the face facial action unit recognition module is used to respectively recognize and process each facial picture in the multiple facial pictures included in the target picture group to determine the action features of each facial action unit in each facial picture; and determine the emotion features of each facial picture according to the action features of each facial action unit in each facial picture; the emotion state parameter calculation module is used to calculate the emotion state parameters of the target user according to the preset processing rules and the emotion features of each facial picture; A second calling module is used to call a preset transaction risk identification model to process the emotional state parameters of the target user to obtain a corresponding target processing result; A determination module is used to determine whether the transaction behavior in which the target user participates has a preset transaction risk according to the target processing result; the preset transaction risk includes: fraud risk; Among them, the first calling module is specifically used to use the face facial action unit recognition module to determine each face facial action unit in each face picture in the target picture group; and extract the action features of each face facial action unit in each face picture; based on the preset mapping relationship, determine the emotional features of each face facial action unit according to the action features of each face facial action unit; respectively integrate the emotional features of multiple face facial action units in each face picture, and obtain the emotional features of each face picture through weighted operation; use the emotional state parameter calculation module according to the preset processing rules, along the time axis, by performing time point-based integration processing on the emotional features of multiple face pictures arranged in chronological order, to obtain the emotional state parameters that can reflect the emotional changes of the target user within a time period; wherein, the preset processing rule is a calculation rule based on integration; Among them, the device is also used to display relevant risk warning information to the target user after determining that the transaction behavior in which the target user participates has a preset transaction risk based on the target processing result; and obtain descriptive data about the transaction behavior provided by the target user; wherein the descriptive data includes descriptive data about the content of the behavior and the cause of the behavior; call a pre-selected and trained semantic recognition model to process the descriptive data to obtain a corresponding semantic recognition result; based on the semantic recognition result, confirm whether the transaction behavior currently participated by the target user really has a preset transaction risk; if it is confirmed that the transaction behavior currently participated by the target user really has a preset transaction risk, the target user is dissuaded accordingly by notifying nearby staff, or the target user is interrupted by interrupting the network connection of the terminal device on which the user currently relies when performing the transaction.

10. A server, characterized in that: The method comprises a processor and a memory for storing processor-executable instructions, wherein the processor implements the steps of the method according to any one of claims 1 to 8 when executing the instructions.

11. A computer-readable storage medium, characterized in that: Computer instructions are stored thereon, and when the instructions are executed, the steps of the method according to any one of claims 1 to 8 are implemented.

Images