Interaction control method, intelligent terminal, storage medium and product

Abstract

The invention discloses an interaction control method, an intelligent terminal, a storage medium and a product, and relates to the technical field of communication security, and the method comprises the steps: collecting interaction information in an interaction scene; performing risk assessment on the interaction scene based on the interaction information through a preset risk assessment model; and under the condition of determining that the interaction scene has the risk, the intelligent terminal executes a corresponding defense measure according to a risk assessment result. The fraud or harassment risk identification precision can be significantly improved, intelligent early warning and personalized defense can be realized in combination with personalized early warning measures, the user experience and safety are improved, and a safe, efficient and personalized communication environment is provided for users.

Description

Technical Field

[0001] The present application relates to the field of communication security, and in particular to an interactive control method, an intelligent terminal, a storage medium and a product. Background Art

[0002] Telecom fraud and harassing calls are becoming increasingly rampant and the situation is grim. Existing telecom fraud and harassing call defense technologies mostly rely on simple keyword filtering, blacklist matching and other methods. These methods have limited recognition accuracy and are prone to false positives and missed positives in the face of diversified fraud methods and hidden harassing behaviors. In addition, existing methods lack in-depth mining and analysis of user behavior characteristics, and cannot provide personalized services based on individual differences of users, resulting in poor user experience. Summary of the invention

[0003] The main purpose of this application is to provide an interactive control method, an intelligent terminal, a storage medium and a product, aiming to provide users with a safe, efficient and personalized communication environment.

[0004] To achieve the above purpose, the present application proposes an interactive control method, which is applied to a smart terminal, and the method includes:

[0005] Collect interaction information in interactive scenarios;

[0006] Based on the interaction information, and using a preset risk assessment model, a risk assessment is performed on the interaction scenario;

[0007] When it is determined that the interaction scenario has risks, the smart terminal executes corresponding defense measures according to the risk assessment result.

[0008] In one embodiment, the step of performing risk assessment on the interaction scenario based on the interaction information and using a preset risk assessment model includes:

[0009] Extracting features from the interaction information to obtain key feature information, where the key feature information includes: one or more of voice features, text features, and picture features;

[0010] Converting the key feature information into a key feature vector;

[0011] The key feature vector is input into a preset risk assessment model, the key feature vector is analyzed and identified through the risk assessment model, and the fraud or harassment risk of the interaction scenario is assessed based on the analysis and identification results.

[0012] In one embodiment, the step of analyzing and identifying the key feature vector by using the risk assessment model, and performing fraud or harassment risk assessment on the interaction scenario according to the analysis and identification result includes:

[0013] By using the risk assessment model, combined with the pre-built fraud or harassment feature library and anti-fraud or anti-harassment feature library, the key feature vector is analyzed and identified to obtain an initial analysis and identification result;

[0014] The initial analysis and identification results are cross-validated in combination with the pre-built cloud database and local database, and the fraud or harassment risk of the interaction scenario is assessed based on the cross-validation results.

[0015] In one embodiment, before the step of combining the pre-built cloud database and the local database to cross-validate the initial analysis and identification results, the step further includes:

[0016] When constructing the cloud database, desensitization processing is performed on the data in the cloud database.

[0017] In one embodiment, the risk assessment model training step includes:

[0018] Building a fraud or harassment feature library, an anti-fraud or anti-harassment feature library based on historical interaction data and / or constructed interaction data;

[0019] Using the fraud or harassment feature library and the anti-fraud or anti-harassment feature library, model training is performed based on a deep neural network framework to obtain an initial deep neural network model;

[0020] Based on a preset tuning strategy, the parameters of the initial deep neural network model are tuned to obtain a preset risk assessment model.

[0021] In one embodiment, the fraud or harassment feature library includes: one or more features of impersonation, creation of emergency or tempting situations, professional rhetoric and techniques, and request for sensitive information or transfer of funds; the anti-fraud or anti-harassment feature library includes: one or more features of social network, timbre, accent, language habits, and social object portraits.

[0022] In one embodiment, when it is determined that the interaction scenario has a risk, the step of the smart terminal executing corresponding defense measures according to the risk assessment result includes:

[0023] In the case where it is determined that there is a risk of fraud or harassment in the interaction scenario, the smart terminal calls a pre-defined defense measure corresponding to the user in the current interaction scenario according to the risk assessment result;

[0024] Execute the defense measures corresponding to the user in the current interaction scenario, or execute corresponding defense measures based on the priority of the defense measures in the current interaction scenario.

[0025] In one embodiment, the interaction scenario includes: one or more of a voice call scenario, a text chat scenario, and a video interaction scenario; the defense measures include: interrupting the current interaction, voice or text message reply, adding the other party's number to a blacklist or reporting; the priorities include: reminder, warning, intervention, and interruption.

[0026] In one embodiment, when it is determined that the interaction scenario has a risk, before the step of the smart terminal executing corresponding defense measures according to the risk assessment result, the step further includes:

[0027] Develop corresponding defense measures for users through user behavior pattern mining models.

[0028] In one embodiment, the step of formulating corresponding defense measures for users through the user behavior pattern mining model includes:

[0029] Through the user behavior pattern mining model, the user behavior characteristics in the corresponding interaction scenarios in the historical interaction data are mined and analyzed to obtain the user interaction behavior pattern;

[0030] Based on the user behavior characteristics and user interaction behavior patterns, construct a user behavior feature library;

[0031] Corresponding defense measures are formulated for the user based on the user behavior feature library.

[0032] In one embodiment, the step of mining and analyzing the user behavior characteristics in the corresponding interaction scenarios in the historical interaction data by using the user behavior pattern mining model includes:

[0033] By using the user behavior pattern mining model, mining the interaction behavior features in the corresponding interaction scenarios in the historical interaction data, wherein the interaction behavior features include: one or more of voice features, text features, and picture features;

[0034] The interactive behavior characteristics are analyzed from multiple dimensions to obtain a user interactive behavior pattern, wherein the multiple dimensions include: one or more of interaction duration, interaction frequency, and temporal communication relationship.

[0035] In one embodiment, the method further comprises:

[0036] Collect new interaction data, and based on the new interaction data, update the parameters of the risk assessment model to obtain an updated risk assessment model; and / or update the user behavior feature library based on the new interaction data and user behavior features.

[0037] In one embodiment, the intelligent terminal is configured with an anti-fraud decision-making agent, and the step of performing risk assessment on the interaction scenario based on the interaction information and through a preset risk assessment model includes:

[0038] The anti-fraud decision-making agent calls a preset risk assessment model based on the interaction information to perform risk assessment on the interaction scenario, and outputs defense measures if it is assessed that there is a risk in the interaction scenario.

[0039] The present application also proposes an interactive control device, the device comprising:

[0040] A collection module, used to collect interaction information in an interaction scenario;

[0041] An assessment module, configured to perform risk assessment on the interaction scenario based on the interaction information and through a preset risk assessment model;

[0042] The defense processing module is used to execute corresponding defense measures according to the risk assessment result when it is determined that the interaction scenario has risks.

[0043] The present application also proposes an intelligent terminal, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the interactive control method described above.

[0044] The present application also proposes a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the interactive control method described above are implemented.

[0045] The present application also proposes a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps of the interactive control method described above are implemented.

[0046] One or more technical solutions proposed in this application have at least the following technical effects: by collecting interactive information in interactive scenarios; based on the interactive information and through a preset risk assessment model, risk assessment is performed on the interactive scenario; when it is determined that there is a risk in the interactive scenario, the intelligent terminal executes corresponding defense measures according to the risk assessment results. This application uses a risk assessment model to conduct risk assessment on interactive scenarios, which can significantly improve the accuracy of identifying fraud or harassment risks. At the same time, combined with personalized defense measures, it can achieve intelligent early warning and personalized defense, improve user experience and security, and thus provide users with a safe, efficient, and personalized communication environment. This application has broad application prospects and important social value in the field of telecommunications fraud and harassment call defense. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0048] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0049] Figure 1 A schematic diagram of a flow chart provided for the first embodiment of the interactive control method of the present application;

[0050] Figure 2 A flowchart of the second embodiment of the interactive control method of the present application is provided;

[0051] Figure 3 A flowchart of the third embodiment of the interactive control method of the present application is provided;

[0052] Figure 4 A flowchart of the fourth embodiment of the interactive control method of the present application is provided;

[0053] Figure 5 A complete flow chart of the interactive control method provided in the embodiment of the present application;

[0054] Figure 6 This is a schematic diagram of the module structure of the interactive control device according to an embodiment of the present application;

[0055] Figure 7 A schematic diagram of the structure of an intelligent terminal of the hardware operating environment involved in the interactive control method in the embodiment of the present application.

[0056] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0057] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0058] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0059] The main solution of the embodiment of the present application is: by collecting interactive information in interactive scenarios; based on the interactive information and through a pre-trained deep neural network model, the interactive scenario is risk assessed; when it is assessed that the interactive scenario is risky, the interactive scenario is defended based on personalized early warning measures. The present application uses a deep neural network model to conduct risk assessments on interactive scenarios, which can significantly improve the accuracy of identifying fraud or harassment risks. At the same time, combined with personalized early warning measures, it can achieve intelligent early warning and personalized defense, improve user experience and security, and thus provide users with a safe, efficient, and personalized communication environment. The present application has broad application prospects and important social value in the field of telecommunications fraud and harassment call defense.

[0060] Existing telecommunications fraud and nuisance call defense technologies mostly rely on simple keyword filtering, blacklist matching and other methods. These methods have limited recognition accuracy and are prone to false positives and missed positives in the face of diversified fraud methods and hidden nuisance behaviors. In addition, existing methods lack in-depth mining and analysis of user behavior characteristics, and are unable to provide personalized services based on individual user differences, resulting in poor user experience.

[0061] This application takes into account that in recent years, the rapid development of technologies such as artificial intelligence, big data, and cloud computing has brought new opportunities to the field of telecommunications anti-fraud. In particular, the rise of deep neural network (DNN) technology enables the system to identify and prevent telecommunications fraud more quickly and accurately. At the same time, user behavior pattern mining (UBPM) technology also provides strong support for the formulation of personalized defense strategies.

[0062] Therefore, the present application provides an intelligent anti-telecom fraud and harassing call comprehensive defense system and method that integrates deep neural network (DNN) and user behavior pattern mining (UBPM) technology, aiming to build a safe, efficient and personalized communication environment for users through high-precision identification, real-time warning, personalized defense strategies and continuous learning and optimization mechanisms.

[0063] The system architecture involved in the solution of this embodiment may include a client and / or a server, and may further include a storage device, a network interface, an operating system, a database system, a programming language and framework, middleware and API, etc. Take the system architecture including a client and a server as an example, where:

[0064] The client is used to provide users with an interactive interface, such as a call interface, a chat interface (including a text or audio and video interface), etc., and can perform AI recognition, AI reply, AI interception and other operations in the interactive interface based on the defense strategy feedback from the server.

[0065] As the background, the server can deploy large models and process the client's interaction data, such as identifying and assessing the risks of interaction information.

[0066] In this embodiment, the server can be a high-performance server cluster, and can deploy preset risk assessment models and user behavior pattern mining (UBPM) models such as deep neural network (DNN) models. The server should have strong computing power, storage capacity and network bandwidth to ensure that the system can process a large amount of call data in real time.

[0067] Storage devices: used to store key information such as user call data, DNN model parameters, UBPM feature library, etc. Storage devices have high reliability, high availability and high scalability to ensure data integrity and stable operation of the system.

[0068] Network interface: used to connect the system to other communication networks, such as the telecommunications operator's network, the Internet, etc. The network interface should be high-speed, stable, and low-latency to ensure that the system can obtain and process call data in real time.

[0069] Operating system: You can choose a stable, secure and easy-to-maintain operating system, such as Linux or Windows Server. The operating system should have strong resource management, process scheduling and security protection capabilities.

[0070] Database system: used to store and manage key data such as user behavior feature database, blacklist, warning records, etc. The database system should have high performance, high availability and high security to ensure data integrity and stable operation of the system. Cloud database and local database can be configured according to data storage requirements.

[0071] Programming language and framework: You can choose efficient, stable and easily extensible programming languages, such as Python, Java, etc. At the same time, use deep learning frameworks (such as TensorFlow, PyTorch, etc.) and machine learning libraries (such as scikit-learn, etc.) to build and train DNN models, as well as implement data mining and machine learning algorithms for the UBPM system.

[0072] Middleware and API: A rich set of middleware and API interfaces can be provided, such as graphical interfaces or video interfaces, to achieve integration and interaction between the system and other applications or services. For example, communication with telecom operators can be achieved through RESTful API interfaces, and real-time warning notifications can be achieved through WebSocket.

[0073] Based on the above system architecture, the present application embodiment provides an interactive control method, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the interactive control method of the present application.

[0074] In this embodiment, the interactive control method includes steps S10 to S30, each of which is explained in detail below.

[0075] like Figure 1 As shown, the first embodiment of the present application proposes an interactive control method, which is applied to a smart terminal. The method includes:

[0076] Step S10, collecting interaction information in the interaction scenario;

[0077] The solution of this embodiment is mainly applied to smart terminals that can provide interactive functions for users. Interactive software or programs can be set on the smart terminals, and corresponding client interfaces can be provided, such as call interfaces, chat interfaces (including text or audio and video interfaces), etc.

[0078] The technical solution of this embodiment can integrate machine learning model algorithms such as deep neural networks and user behavior pattern mining technology, adopt an intelligent comprehensive defense method against telecommunications fraud and harassing calls, and build a safe, efficient and personalized communication environment for users through high-precision identification, real-time warning, personalized defense measures and continuous learning and optimization mechanisms.

[0079] The technical solution of this application is mainly used in interactive scenarios where fraud or harassment exists, such as text messages, phone calls, instant messaging (such as WeChat, QQ), Tik Tok and other voice or video call (or conversation) scenarios.

[0080] The execution subject of this embodiment may be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone and other smart terminals, or an electronic device capable of implementing the above functions, a server, etc. The following takes a smart terminal as an example to explain the solution of this embodiment.

[0081] First, when the user is in an interactive scene, the interaction information of the user in the interactive scene is collected.

[0082] The interaction scenarios may include: one or more of voice call scenarios, text chat scenarios, and video interaction scenarios. Specifically, the interaction scenarios may be voice or video call (or dialogue) scenarios such as text messages, phone calls, instant messaging (such as WeChat, QQ, etc.), and vibrato. These interaction scenarios may all involve the risk of fraud or harassment.

[0083] Among them, depending on the interaction scenario, the interaction information may include: text, voice, picture, video, etc. By collecting the interaction information of both users in the interaction scenario, such as call voice, the interaction information can be identified, and then the risk assessment of the interaction scenario can be performed so as to take corresponding early warning and defense measures.

[0084] Among them, as an implementation method, after collecting the user's interactive information, the interactive data can be pre-processed by cleaning, denoising, standardization and other operations to improve the data quality and the accuracy of subsequent analysis.

[0085] Step S20, performing risk assessment on the interaction scenario based on the interaction information and through a preset risk assessment model;

[0086] As an implementation method, the intelligent terminal may collect interaction information in an interaction scenario, and perform risk assessment on the interaction scenario based on the interaction information and through a preset risk assessment model.

[0087] As another implementation, the intelligent terminal may send the collected interaction information in the interaction scenario to the server, and the server may perform risk assessment on the interaction scenario based on the interaction information and through a preset risk assessment model.

[0088] Among them, risk assessment of interactive scenarios mainly involves fraud or harassment risk assessment. For example, during a call, based on the call content of the two parties, it is assessed whether there is a fraud risk in the current call scenario; for example, in a chat scenario, based on the chat content, it is assessed whether there is a fraud risk in the current chat scenario.

[0089] Among them, the preset risk assessment model can be constructed based on a variety of machine learning algorithms and / or big data analysis. These algorithms can be, for example, LightGBM, XGBoost, random forest, logistic regression, GBDT, neural network, ensemble learning and other commonly used model algorithms. This embodiment takes the construction of a risk assessment model based on deep neural networks (DNN) as an example.

[0090] Among them, deep neural networks are a type of neural network with multiple nonlinear transformations. These neural networks can learn complex patterns and representations of data by simulating the way the human brain processes information. The key feature of deep neural networks is "depth", that is, they contain multiple hidden layers, which enables the network to learn high-level abstractions and features of data.

[0091] The model training principle of deep neural network involves several key steps, including forward propagation, loss function calculation, back propagation and parameter update. The following is a detailed description of these steps:

[0092] Forward Propagation:

[0093] Input data is fed into the input layer of the neural network, and the data passes through each layer of neurons, each of which performs a weighted sum on the data and applies an activation function. This process continues until the data reaches the output layer, where the neurons give the prediction results.

[0094] Loss Function calculation:

[0095] The loss function measures the difference between the model's predicted value and the true value. Common loss functions include mean squared error (MSE) for regression problems and cross entropy loss for classification problems. The value of the loss function is used to evaluate the performance of the model and serves as the optimization target.

[0096] Backpropagation:

[0097] Backpropagation is a method of propagating an error signal through a neural network in order to calculate the gradient of the loss function with respect to each parameter (weight and bias). This process starts at the output layer and goes backwards through each layer of the network, using the chain rule to calculate the gradient. The gradient represents the rate of change of the loss function with respect to the parameters, indicating how to adjust the parameters to reduce the loss.

[0098] Parameter Update:

[0099] Use an optimization algorithm (such as gradient descent) to update the network's weights and biases based on the calculated gradient. The optimization algorithm determines the update step size (learning rate) for each parameter to ensure that the model moves in the downward direction of the loss function. This process is repeated many times until the model's performance on the training data reaches a satisfactory level or a predetermined number of iterations is reached.

[0100] The model training process is iterative, which means that the above steps will be repeated many times until the performance of the model on the training set is no longer significantly improved or the preset stopping condition is reached.

[0101] In addition, during the model training process, overfitting and regularization are also involved. Overfitting refers to the situation where the model performs well on the training data but performs poorly on unseen data. To prevent overfitting, regularization techniques such as L1, L2 regularization or Dropout can be used. Regularization limits the complexity of the model by adding additional penalty terms to the loss function.

[0102] Finally, the model performance can be evaluated and validated. During the training process, the data is usually divided into training set, validation set and test set. The training set is used to train the model, the validation set is used to evaluate the performance of the model during the training process to adjust the hyperparameters, and the test set is used to evaluate the final performance of the model after the training is completed.

[0103] Among them, early stopping (a technique to prevent overfitting) can be used to monitor the performance on the validation set and decide when to stop training. If the performance on the validation set does not improve in multiple consecutive iterations, the training can be terminated early.

[0104] Through these steps, deep neural networks are able to learn complex data patterns and provide powerful predictive capabilities in a variety of tasks.

[0105] The risk assessment model of this embodiment can be constructed based on the above-mentioned deep neural network (DNN), or can be constructed based on other model algorithms. Specifically, it can include information collection, feature recognition, risk assessment and early warning prompts, etc., so as to identify and assess risks such as fraud by collecting and analyzing data.

[0106] In a specific interaction scenario, after the interaction information of the two parties is collected, the interaction scenario can be risk assessed based on the interaction information and through the above-mentioned preset risk assessment model to obtain a risk assessment result, so as to determine whether to execute corresponding defense measures according to the risk assessment result. Among them, taking the risk assessment model based on a deep neural network as an example, as an implementation method, the risk assessment of the interaction scenario based on the interaction information and through the preset risk assessment model can include the following steps:

[0107] First, feature extraction is performed on the interaction information to obtain key feature information;

[0108] The key feature information may include one or more of voice features, text features, and picture features.

[0109] Then, the key feature information is converted into a key feature vector;

[0110] Finally, the key feature vector is input into a pre-trained deep neural network model, the key feature vector is analyzed and identified by the deep neural network model, and the fraud or harassment risk of the interaction scenario is assessed based on the analysis and identification results.

[0111] Among them, based on key feature information, the user's calling behavior can be deeply analyzed to discover the user's calling habits, preferences, and potential risky behavior patterns.

[0112] Therefore, the key feature vectors can be input into a pre-trained deep neural network model, and the key feature vectors can be analyzed and identified by the deep neural network model, and then the fraud or harassment risk of the interaction scenario can be assessed based on the analysis and identification results.

[0113] Among them, the deep neural network model can be obtained by model training based on a pre-built fraud or harassment feature library and an anti-fraud or anti-harassment feature library.

[0114] Among them, the fraud or harassment feature library can be constructed based on studying public channels or legal cases, or historical interaction data; the anti-fraud or anti-harassment feature library can be obtained through behavioral learning and summary based on the normal telephone and message characteristics of users in historical interaction data.

[0115] Among them, the fraud or harassment feature library may include: one or more features of impersonation, creation of emergency or tempting situations, professional rhetoric and techniques, and requirements for sensitive information or money transfers; the anti-fraud or anti-harassment feature library includes: one or more features of social networks, timbre, accent, language habits, and social object portraits.

[0116] Wherein, as an implementation manner, the step of analyzing and identifying the key feature vector by using the deep neural network model and performing fraud or harassment risk assessment on the interaction scenario according to the analysis and identification result may include:

[0117] By using the deep neural network model, combined with the pre-built fraud or harassment feature library and anti-fraud or anti-harassment feature library, the key feature vector is analyzed and identified to obtain an initial analysis and identification result;

[0118] The initial analysis and identification results are cross-validated in combination with the pre-built cloud database and local database, and the fraud or harassment risk of the interaction scenario is assessed based on the cross-validation results, wherein the data in the cloud database may be desensitized data.

[0119] Among them, the cloud database collects a large amount of various fraud or harassment information in interactive scenarios involving fraud or harassment, including the words, voice, characters, etc. in the user communication process, forming various scripts and common characteristics of fraud or harassment, and provides them to the local for matching evaluation. Among them, for the protection of user privacy, when building the cloud database, the data in the cloud database can be desensitized, that is, the information involving user privacy in the cloud database is desensitized.

[0120] Among them, the local database collects various historical communication records of the current user, such as historical call records, chat records, the gender, voice, language habits, etc. of the caller, and stores the corresponding contacts, call information, conversation information, and character relationships in the local database. Therefore, through matching the local database, it is possible to determine the interaction scenario of the current user and whether he is a known contact, such as family members, relatives, friends, etc. who have call records and are safe.

[0121] Among them, information related to user privacy can be stored in a local database.

[0122] Therefore, by configuring the cloud database and the local database, and combining the cloud database and the local database, the current interaction scenario of the current user can be cross-verified to determine whether there is a risk of fraud or harassment, and the security and reliability of the interaction scenario can be improved.

[0123] For example, as an implementation method, a deep neural network model can be used to first conduct a fraud or harassment risk assessment on the current user's current interaction scenario in combination with a local database to determine whether the current call partner is a known contact, such as a relative, friend, or other safe contact with whom a call record has been made. If the current call partner is determined not to be a known contact in combination with the local database, a large amount of fraud or harassment information collected in a cloud database in various interaction scenarios involving fraud or harassment can be further combined to conduct a fraud or harassment risk assessment on the current user's current interaction scenario, thereby more accurately identifying fraud and harassment risks and building a safe communication environment for users.

[0124] Specifically, taking a call scenario as an example, if the current user receives a call, before the call, the system can identify whether the incoming call number is a known number in the current user's address book. If so, the call is allowed without a risk warning. If it is not a known number, the system can identify whether the incoming call number is a marked fraud call or a harassment call (which can be identified in conjunction with an existing anti-fraud APP). If so, it prompts that the current number is a fraud call or a harassment call. If not, the call can be made normally.

[0125] During a call between two users, first, the voice timbre, pitch, tone, language habits and call content of the current user's call partner can be collected, and the collected information can be matched with the local database to determine whether the call partner is a known contact of the current user (such as a relative, friend, etc.).

[0126] If it is determined in combination with the local database that the current call recipient is not a known contact, the call content of the current user can be further verified by combining a large amount of fraud or harassment information collected in the cloud database in interactive scenarios involving fraud or harassment. This can more accurately identify the risks of telecommunications fraud and harassment and build a safe communication environment for users.

[0127] In addition, after the local call ends, the content of the call can be used as known data and given to the deep neural network model for re-training to improve the accuracy and reliability of the deep neural network model.

[0128] Step S30: When it is determined that the interaction scenario has risks, the smart terminal executes corresponding defense measures according to the risk assessment result.

[0129] In this embodiment, corresponding personalized defense measures are pre-formulated for risk interaction scenarios of different users, and the personalized defense measures may include various personalized defense strategies.

[0130] As an implementation method, corresponding defense measures can be formulated for users through user behavior pattern mining models.

[0131] For example, a user behavior feature library can be built in advance based on the user behavior pattern mining model. According to the pre-built user behavior feature library, personalized defense measures can be customized for each user, such as automatic hang-up, voice or SMS reply, warning, etc.

[0132] In a specific interaction scenario, for example, if a potential fraud or harassing call is identified in the current call scenario, the early warning mechanism will be triggered immediately, the user will be notified in a variety of ways, and emergency handling suggestions will be provided.

[0133] In addition, personalized defense measures may include a variety of emergency handling suggestions, which can be provided to users simultaneously or in priority order. For example, according to the four-level personalized security warning of reminder-warning-intervention-interruption, an AI processing response warning mechanism is provided for fraudulent calls based on the degree of certainty of identification.

[0134] Specifically, as an implementation method, when it is determined that there is a risk of fraud or harassment in the current interaction scenario, the smart terminal calls the pre-formulated personalized defense measures corresponding to the user in the current interaction scenario based on the risk assessment results, and executes the defense measures corresponding to the user in the current interaction scenario.

[0135] Among them, as an implementation method, when the priority of the defense measures is configured, the smart terminal can execute the corresponding defense measures based on the priority of the defense measures in the current interaction scenario when executing the defense measures corresponding to the user in the current interaction scenario.

[0136] As mentioned above, the interaction scenario may include: one or more of a voice call scenario, a text chat scenario, and a video interaction scenario.

[0137] Correspondingly, the defense measures may include: interrupting the current interaction, voice or SMS reply, adding the other party's number to a blacklist or reporting one or more; the priority of the defense measures may include: reminder, warning, intervention, interruption, etc.

[0138] In specific implementation, the current interaction, voice or SMS reply, blacklisting or reporting of the other party's number can be interrupted based on the personalized defense measures; or, defense processing can be performed based on the priority set for the personalized defense measures.

[0139] Compared with the existing technology, the technical solution of this application can deeply explore the behavioral characteristics of the two parties in the call, combine deep learning algorithms for accurate identification, and effectively deal with telecommunications fraud and harassing calls through personalized defense measures.

[0140] This embodiment collects interactive information in interactive scenarios through the above scheme; based on the interactive information and through a preset risk assessment model, the interactive scenario is risk assessed; when it is determined that the interactive scenario is risky, the intelligent terminal executes corresponding defense measures according to the risk assessment results. This application uses deep neural network models and other methods to conduct risk assessments on interactive scenarios, which can significantly improve the accuracy of identifying fraud or harassment risks. At the same time, combined with personalized early warning measures, it can achieve intelligent early warning and personalized defense, improve user experience and security, and thus provide users with a safe, efficient, and personalized communication environment. This application has broad application prospects and important social value in the field of telecommunications fraud and harassment call defense.

[0141] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the first embodiment can refer to the above introduction, and will not be repeated in the following. Figure 2 Before step S10, the interactive control method further includes step S100.

[0142] Step S100, training a risk assessment model.

[0143] Among them, the risk assessment model can be constructed based on a variety of machine learning algorithms and / or big data analysis, such as LightGBM, XGBoost, random forest, logistic regression, GBDT, neural network, ensemble learning and other commonly used model algorithms.

[0144] This embodiment takes the construction of a risk assessment model based on a deep neural network (DNN) as an example.

[0145] As an implementation method, the training process of the risk assessment model may include steps A to C:

[0146] Step A, constructing a fraud or harassment feature library, an anti-fraud or anti-harassment feature library based on historical interaction data and / or constructed interaction data;

[0147] Among them, depending on the interaction scenarios, historical interaction data may include historical call data, text chat data, audio and video interaction information, such as voice data, text data, and call metadata, etc. It may also include online chat data, such as WeChat, Douyin and other audio and video data.

[0148] Among them, historical interaction data can be reasonably collected through a variety of public channels, such as various network big data, police public information, user shared data, user tag data, reasonable data provided by telecommunications operators, etc.

[0149] Among them, as an implementation method, after collecting the historical interaction data, the historical interaction data can be pre-processed by cleaning, denoising, standardization and other operations to improve the data quality and the accuracy of subsequent analysis.

[0150] The constructed interaction data can be artificially constructed according to the characteristics of fraudulent or harassing information, or constructed through a large model.

[0151] Then, a fraud or harassment feature library and an anti-fraud or anti-harassment feature library may be constructed based on the historical interaction data and / or the constructed interaction data.

[0152] The fraud or harassment feature library may include one or more features of: impersonation, creation of emergency or tempting situations, professional rhetoric and techniques, and request for sensitive information or transfer of funds.

[0153] The anti-fraud or anti-harassment feature library may include: one or more features of social relationship network, timbre, accent, language habits, and social object portrait.

[0154] Specifically, taking the fraud feature library as an example, as an implementation method, the construction and learning of the fraud feature library can adopt the following scheme:

[0155] By studying public channels or legal cases, we train the fraud feature library of open channels, that is, the existing fraud calls and information features, and form a knowledge base through behavioral learning and summary. In addition, we continuously conduct learning and summary during actual calls, extract typical features, and use this to supplement the fraud feature library. Currently, the main fraud types include the following types. By analyzing and mining the caller information, we can identify whether they meet the typical types in the fraud feature library.

[0156] 1. Identity fraud

[0157] Impersonating a leader or company boss: calling the person in charge of a grassroots unit or financial personnel, and asking the deceived unit to pay the order amount, handling fee or transfer the money to a designated account under the pretext of selling books, commemorative coins, or project payments.

[0158] Impersonating relatives and friends: Using Trojan programs to steal the other party’s online communication tool password, intercepting the other party’s chat video data, and then impersonating the owner of the communication account to defraud their relatives and friends in the name of emergency matters such as "serious illness, car accident, etc."

[0159] Impersonating public security, procuratorial and judicial personnel: calling the victim's phone and demanding that the victim's funds be transferred to a state account to cooperate with the investigation on the grounds that the victim's identity information has been stolen and that he or she is suspected of crimes such as money laundering and drug trafficking.

[0160] Impersonating staff members: such as those in education, civil affairs, the Disabled Persons’ Federation, medical insurance, social security, banks, radio and television, Taobao customer service, etc., and deceiving victims into providing personal information or transferring money on the grounds of receiving subsidies, relief funds, scholarships, account abnormalities, refunds due to out-of-stock goods, etc.

[0161] 2. Create an urgent or tempting situation

[0162] Fabricating emergencies: For example, fabricating unexpected events such as car accidents, kidnappings, surgeries, etc. to scare the victims and make them anxious to transfer money to solve the problem.

[0163] Providing attractive opportunities: For example, impersonating a well-known company or a popular program production team, sending false winning information to the victim, and inducing them to pay personal income tax, notarization fees, deposits and other fees in order to receive the prize.

[0164] Exploiting curiosity: Sending links or QR codes containing Trojan viruses to trick victims into clicking or scanning in order to steal personal information or funds.

[0165] 3. Use professional words and skills

[0166] Tentative questions: such as "Guess who I am" to induce the victim to identify himself with the person he is dealing with.

[0167] Imitating the tone of acquaintances: By lurking in social platforms such as QQ for a long time to understand the tone of the victim’s relatives, friends or colleagues, and then impersonating their identities to commit fraud.

[0168] Creating a sense of authority: Impersonating staff from public security, judicial, banking and other departments, using professional terminology and a serious tone to create a sense of authority in order to gain trust.

[0169] Exploiting psychological weaknesses: For example, taking advantage of the victim’s eagerness to get rid of the connection and reduce losses, inducing them to transfer funds to a so-called safe account.

[0170] 4. Requesting sensitive information or transferring money

[0171] Request personal information: such as bank card number, password, verification code, etc., in order to steal funds.

[0172] Requesting transfer or remittance: requiring the victim to transfer funds to a designated account for various reasons, such as paying handling fees, deposits, taxes, etc.

[0173] Correspondingly, the following solutions can be adopted for the construction and learning of anti-fraud feature libraries:

[0174] Based on the user's normal phone and message characteristics, a knowledge base is formed through behavioral learning and summarization, and learning and summarizing are continuously carried out during the actual call process to extract the characteristics of calls with acquaintances. That is, through the construction and learning of the anti-fraud feature library, the system will know the user portraits of frequently used call objects centered on the current user, and can perform identity calibration and call security feedback more effectively.

[0175] For example, by continuously learning and accumulating contacts or call titles, a social relationship library centered on the person can be formed; by learning the timbre and accent of contacts or callers, a timbre feature library of frequently used contacts can be formed; by learning the communication or information editing user characteristics and habits of contacts or callers, a communication feature library of frequently used contacts can be formed. Thus, an anti-fraud feature library can be constructed based on the above feature libraries.

[0176] Step B, performing model training based on a deep neural network framework using the fraud or harassment feature library and the anti-fraud or anti-harassment feature library to obtain an initial deep neural network model;

[0177] Among them, a deep neural network model can be built based on a deep neural network framework, such as TensorFlow or PyTorch.

[0178] Among them, when training the deep neural network model, the interactive data in the constructed fraud or harassment feature library and the anti-fraud or anti-harassment feature library can be pre-processed by cleaning, denoising, standardization and other operations to improve data quality and the accuracy of subsequent analysis.

[0179] Then, the deep neural network model is trained using the preprocessed historical interaction data (such as historical call data), and the model parameters are adjusted to optimize the recognition performance.

[0180] As mentioned earlier, deep neural networks are a class of neural networks with multiple nonlinear transformations that are able to learn complex patterns and representations of data by simulating the way the human brain processes information. The key feature of deep neural networks is "depth", that is, they contain multiple hidden layers, which enables the network to learn high-level abstractions and features of the data.

[0181] Among them, the model training principle of deep neural network involves several key steps, mainly including forward propagation, loss function calculation, back propagation and parameter update.

[0182] During the training process, the data is usually divided into training set, validation set and test set. The training set is used to train the model, the validation set is used to evaluate the performance of the model during the training process to adjust the parameters, and the test set is used to evaluate the final performance of the model after the training is completed.

[0183] The model training process is iterative, that is, the above steps are repeated many times until the performance of the model on the training set is no longer significantly improved, or the preset stopping condition is reached.

[0184] Step C: Based on a preset tuning strategy, the parameters of the initial deep neural network model are tuned to obtain a trained deep neural network model.

[0185] Among them, the preset tuning strategy can be selected or configured according to the actual situation. For example, the model parameters can be tuned through cross-validation, grid search and other methods to ensure the stability and accuracy of the model in different scenarios.

[0186] Taking cross validation as an example, the following scheme can be used to tune the model parameters through cross validation method:

[0187] As mentioned above, you can configure a cloud database and a local database. By configuring the cloud database and the local database, and combining the cloud database and the local database, you can cross-validate the initial deep neural network model, and then you can tune the parameters according to the cross-validation results.

[0188] Among them, the cloud database collects a large amount of various fraud or harassment information in interactive scenarios involving fraud or harassment, including the words, voice, characters, etc. during user communication, forming various scripts and common characteristics of fraud or harassment, and provides them to the local for matching and evaluation. Among them, in order to protect user privacy, the cloud database has desensitized the information involving user privacy.

[0189] The local database collects various historical communication records of the current user, such as historical call records, chat records, the gender, voice, language habits, etc. of the caller, and stores the corresponding contacts, call information, conversation information, and character relationships in the local database. By matching the local database, it is possible to determine the interaction scenario of the current user and whether the user is a known contact, such as family members, relatives, friends, etc. who have call records and are safe.

[0190] Among them, information related to user privacy can be stored in a local database.

[0191] After obtaining the initial deep neural network model through model training based on the deep neural network framework through the fraud or harassment feature library and the anti-fraud or anti-harassment feature library, the initial deep neural network model can be further cross-validated in combination with the cloud database and the local database, and the parameters can be further tuned based on the cross-validation results. This can improve the risk identification accuracy of the deep neural network model, so that the deep neural network model can more accurately identify fraud and harassment risks and build a safe communication environment for users.

[0192] It should be noted that the interaction data in the above-mentioned cloud database and local database can also be integrated into the fraud or harassment feature library and the anti-fraud or anti-harassment feature library, so as to enrich the data breadth and depth of the feature library, thereby improving the risk identification accuracy of the deep neural network model.

[0193] This embodiment uses the above scheme to obtain a deep neural network model based on a fraud or harassment feature library, an anti-fraud or anti-harassment feature library, and a neural network framework training to collect interactive information in interactive scenarios; based on the interactive information and through a preset deep neural network model, the interactive scenario is risk assessed; when it is determined that the interactive scenario is risky, the smart terminal executes corresponding defense measures according to the risk assessment results. This application uses a deep neural network model to conduct risk assessments on interactive scenarios, which can significantly improve the recognition accuracy of fraud or harassment risks. At the same time, combined with personalized early warning measures, it can achieve intelligent early warning and personalized defense, improve user experience and security, and thus provide users with a safe, efficient, and personalized communication environment.

[0194] Based on the second embodiment of the present application, in the third embodiment of the present application, the same or similar contents as those of the second embodiment can be referred to the above introduction, and will not be repeated in the following. Figure 3 In step S30, when it is determined that the interaction scenario has a risk, the smart terminal executes corresponding defense measures according to the risk assessment result before ( Figure 3 Taking the example before step S100), the interactive control method also includes step S101.

[0195] Step S101, formulate corresponding defense measures for users through user behavior pattern mining model.

[0196] Specifically, as an implementation mode, step S101, formulating corresponding defense measures for users through user behavior pattern mining models may include: steps S1011 to S1013.

[0197] Step S1011, mining and analyzing the user behavior characteristics in the corresponding interaction scenarios in the historical interaction data through a user behavior pattern mining model to obtain a user interaction behavior pattern;

[0198] Step S1012, constructing a user behavior feature library based on the user behavior features and user interaction behavior patterns;

[0199] Step S1013: formulate corresponding defense measures for the user based on the user behavior feature library.

[0200] Among them, user behavior characteristics may include key characteristics such as interaction duration, interaction frequency, and the current user's interaction object in the current interaction scenario.

[0201] For example, taking a call scenario as an example, user behavior characteristics may include key features such as call duration, call frequency, and call object.

[0202] In order to construct a user behavior feature library, user behavior features can be obtained based on the collected historical interaction data.

[0203] In this embodiment, the user behavior characteristics in the historical interaction data may be mined and analyzed through a user behavior pattern mining model.

[0204] As an implementation method, the step of mining and analyzing the user behavior characteristics in the corresponding interaction scenarios in the historical interaction data through the user behavior pattern mining model may include:

[0205] By using the user behavior pattern mining model, mining the interaction behavior features in the corresponding interaction scenarios in the historical interaction data, wherein the interaction behavior features include: one or more of voice features, text features, and picture features;

[0206] The interactive behavior characteristics are analyzed from multiple dimensions to obtain a user interactive behavior pattern, wherein the multiple dimensions include: one or more of interaction duration, interaction frequency, and temporal communication relationship.

[0207] Specifically, the user behavior pattern mining model can be implemented based on the user behavior pattern mining technology.

[0208] User behavior pattern mining (UBPM) technology is a technology that focuses on analyzing and identifying user behavior patterns. It involves extracting valuable information from a large amount of user data to predict and understand user behavior. Therefore, based on user behavior pattern mining technology, a user behavior pattern mining model can be constructed. Through the user behavior pattern mining model, the interactive behavior characteristics of the corresponding interactive scenarios in the historical interactive data are mined, such as: voice features, text features, image features, etc., and the interactive behavior characteristics are analyzed from multiple dimensions such as interaction duration, interaction frequency, and time-series communication relationships to obtain user interactive behavior patterns. For example, in-depth analysis of user call behavior can be conducted to mine the user's call habits, preferences, and potential risk behavior patterns, so as to build a user behavior feature library based on user behavior characteristics and user interactive behavior patterns.

[0209] Specifically, in the call scenario, user feature mining and behavior pattern recognition is to analyze the detailed record (CDR) data of user communication behavior to discover the rules and characteristics of user behavior. In the user behavior pattern mining of telecommunications networks, hybrid probability models and feature engineering methods can be used to analyze the call behavior characteristics of users in calls from the perspective of user groups and individuals, including multi-dimensional characteristics such as call duration, call frequency, and time sequence communication relationship, and further refine from different time granularities such as hours, days, and weeks to achieve effective discovery of call behavior patterns of different user groups. By analyzing these characteristics, user behavior characteristics with good distinguishability and computability can be formed, which is of great significance for user group classification and behavior pattern mining.

[0210] Among them, the temporal communication relationship feature refers to the temporal communication relationship feature of the user in different time periods that can be obtained by translation on the time axis. After obtaining the user's temporal communication relationship feature, the frequent calling and frequently called information of the user in different time periods can be obtained through the frequent pattern mining algorithm.

[0211] In addition, considering that different user groups have different calling behavior patterns, the user feature vector is constructed by extracting features such as user call frequency, call duration, and temporal communication relationship: For non-numerical features such as user call frequency distribution and call duration distribution, a Gaussian mixture probability model can be introduced for characterization, in which the model parameters serve as the characteristics of the user's calling behavior distribution.

[0212] Through these methods, we can effectively mine and identify the user's behavior patterns in call scenarios and obtain user behavior characteristics. Based on the user behavior characteristics, we can build a user behavior feature library and formulate personalized defense strategies for users based on the user behavior feature library.

[0213] Taking anti-telecom fraud as an example, through user feature mining and behavior pattern recognition technology, potential fraud behaviors can be identified and prevented, and user behavior characteristics can be collected. In anti-telecom fraud, large-scale telecom fraud-related data is first obtained from multiple sources, and data cleaning and feature engineering are performed to extract and construct meaningful features in preparation for subsequent analysis and modeling. Then, machine learning algorithms such as random forest, XGBoost, and LightGBM can be used to predict fraud behaviors. These algorithms can handle nonlinear relationships and improve overall prediction accuracy. After that, the fraudulent numbers can be detected. The fraudulent numbers are significantly different from normal numbers in terms of call behavior characteristics, such as the number of calls per day, the calling rate, the number of out-of-town calls, the call frequency, the call duration, the callback rate, etc. The phone numbers are classified based on these behavioral characteristics, for example, through machine learning algorithms such as KNN, SVM, and decision trees, to identify and classify, and obtain the user's behavior pattern.

[0214] Among them, sequence pattern mining technology can be used to identify frequent patterns in user behavior. These patterns help detect abnormal behavior patterns and promptly discover fraud, malicious attacks and other behaviors.

[0215] In addition, based on multi-dimensional big data analysis such as fraud scenarios, number behavior characteristics, number features, number activity characteristics, number social networks, behavioral event flows, and regions, different social relationships can be identified from the perspective of call pattern clustering. Through these technologies and methods, valuable information can be effectively mined from user communication records and behavior data, the characteristics and patterns of fraudulent behavior can be identified, and a user behavior feature library can be constructed based on this.

[0216] Then, based on the user behavior feature library, personalized defense measures can be customized for each user, such as automatic hang-up, voice or SMS reply, etc.

[0217] For users or numbers that frequently harass, the system can automatically add them to the blacklist and report them to relevant departments for further processing.

[0218] Among them, personalized defense measures may include a variety of emergency handling suggestions, which can be provided to users at the same time or in priority. For example, according to the four-level personalized security warning of reminder-warning-intervention-interruption, an AI processing response warning mechanism is provided for fraudulent calls based on the degree of certainty of identification.

[0219] Therefore, when a potential fraud or harassing call is identified in a call scenario, the early warning mechanism can be triggered immediately, the user can be notified in a variety of ways, and emergency handling suggestions can be provided, thereby providing users with a safe, efficient and personalized communication environment.

[0220] It should be noted that the user behavior characteristics identified by the above-mentioned user behavior pattern mining technology and the user behavior feature library constructed based on the user behavior characteristics can also be integrated into the fraud or harassment feature library, the anti-fraud or anti-harassment feature library, so as to enrich the data breadth and depth of the model training feature library, thereby improving the risk identification accuracy of the deep neural network model and improving the efficiency and accuracy of anti-telecom fraud.

[0221] In addition, it should be noted that when collecting user data, we will strictly abide by relevant laws and regulations to ensure the privacy of user data. For example, we will use technical means such as data encryption and anonymization to protect sensitive information such as user call content and call recipients.

[0222] Based on the third embodiment of the present application, in the fourth embodiment of the present application, the same or similar contents as those in the third embodiment can refer to the above description, and will not be repeated in the following. Figure 4 In step S30, when it is determined that there is a risk in the interaction scenario, after the smart terminal executes corresponding defense measures according to the risk assessment result, the interaction control method further includes step S40.

[0223] Step S40, collect new interaction data, update the parameters of the risk assessment model based on the new interaction data to obtain an updated risk assessment model; and / or update the user behavior feature library based on the new interaction data and user behavior features.

[0224] This embodiment mainly involves the update of the risk assessment model and the user behavior feature library. A continuous learning and optimization mechanism can be adopted, and an online learning algorithm can be introduced to enable the system to continuously update the risk assessment model and the user behavior feature library mined by user behavior patterns according to newly emerging fraud and harassment methods.

[0225] For example, new user call data can be collected regularly to retrain the deep neural network model to adapt to new fraud and harassment methods.

[0226] For another example, the user behavior feature library can be updated in real time based on the user's latest call data and behavior features.

[0227] In addition, the defense strategy can be continuously optimized through the user feedback mechanism to improve the system's recognition accuracy and user experience.

[0228] In addition, the system performance can be evaluated and optimized regularly or irregularly. For example, the key performance indicators such as the recognition accuracy, warning speed, and user experience of the system can be evaluated regularly, and then the system can be optimized and improved based on the evaluation results to improve the overall performance of the system and user experience.

[0229] As for the user interface of smart terminals, smart terminals support AI intelligent recognition. Users can set up AI to reply, reject, hang up, report, and perform other operations on the user interface for harassment and fraud calls.

[0230] This embodiment uses the above scheme, specifically by collecting interactive information in interactive scenarios; based on the interactive information and through a preset risk assessment model, the interactive scenario is risk assessed; when it is determined that the interactive scenario is risky, risk assessment. This application uses deep neural network models and other methods to conduct risk assessments on interactive scenarios, which can significantly improve the recognition accuracy of fraud or harassment risks. At the same time, combined with personalized early warning measures, it can achieve intelligent early warning and personalized defense, improve user experience and security, and thus provide users with a safe, efficient, and personalized communication environment. This application has broad application prospects and important social value in the field of telecommunications fraud and harassment call defense; in addition, this application can also continuously update the deep neural network model and the user behavior feature library mined by user behavior patterns according to newly emerging fraud and harassment methods, which can more accurately identify fraud and harassment risks, and can continuously optimize defense strategies through user feedback mechanisms to improve the recognition accuracy and user experience of the system.

[0231] Furthermore, an anti-fraud decision-making agent can also be configured on the smart terminal or server. When performing risk assessment on the interaction scenario based on the interaction information and through a pre-trained deep neural network model, the anti-fraud decision-making agent can receive input user interaction information, and based on the interaction information, the anti-fraud decision-making agent can call a preset risk assessment model to perform risk assessment on the interaction scenario, and output defense measures when it is determined that the interaction scenario is risky.

[0232] Among them, the anti-fraud decision-making intelligent agent can be configured separately from the risk assessment model, and the two can also be integrated into one, which is not limited in this embodiment.

[0233] The overall process of this embodiment can refer to Figure 5 shown.

[0234] In the field of artificial intelligence (AI), an agent refers to an entity that can perceive the environment and take actions to achieve specific goals. An agent can be software, hardware or a system with autonomy, adaptability and interaction capabilities. It senses changes in the environment (such as sensing the state of its environment through sensors or data input), processes the sensed information, and makes judgments and decisions based on certain decision-making mechanisms (such as based on the knowledge and algorithms learned by itself), and then performs actions to affect the environment or achieve predetermined goals. The core of an agent is that it can continuously learn and optimize its own decision-making and action strategies through interaction with the environment, adapt to changes in the environment through autonomous learning and continuous evolution, and adjust its own behavior and strategies to maintain optimal performance, better complete tasks and adapt to complex environments.

[0235] The intelligent agent used in this application can be a model-based learning intelligent agent, whose working principle involves the key steps of perceiving the environment, making decisions and executing actions, and learning optimization strategies through algorithms to achieve goals.

[0236] In the field of anti-fraud, model-based intelligent agents mainly refer to models built using artificial intelligence technologies such as machine learning. These models can analyze and identify potential fraudulent behaviors. For example, through the application of artificial intelligence technology, the ability to identify fraud gangs can be improved, which will help promote the work of anti-telecom fraud. Through cluster analysis, discrete single-point reports can be clustered into suspicious fraud gangs, and relevant clues can be output to improve the efficiency of telecommunications fraud prevention and governance. Scarce resources such as telephone numbers, application accounts, website domain names, etc. are analyzed, and the different characteristics of resources used for fraud and normally used resources are paid attention to. Machine learning algorithms are used for classification to identify unknown high-risk fraud accounts.

[0237] Through continuous learning and optimization, these model-based agents can improve the efficiency and accuracy of anti-fraud, protect the legitimate rights and interests of users, and provide users with a safe, efficient, and personalized communication environment.

[0238] In addition, to achieve the above purpose, the present application also proposes an interactive control device, such as Figure 6 As shown, the device comprises:

[0239] A collection module 10, used to collect interaction information in an interaction scenario;

[0240] An evaluation module 20, configured to perform risk evaluation on the interaction scenario based on the interaction information and through a preset risk evaluation model;

[0241] The defense processing module 30 is used to execute corresponding defense measures according to the risk assessment result when it is determined that the interaction scenario has risks.

[0242] The interactive control device provided by the present application adopts the interactive control method in the above embodiment, which can significantly improve the recognition accuracy of fraud or harassment risks. At the same time, combined with personalized early warning measures, it can achieve intelligent early warning and personalized defense, improve user experience and security, and thus provide users with a safe, efficient and personalized communication environment. Compared with the prior art, the beneficial effects of the interactive control device provided by the present application are the same as the beneficial effects of the interactive control method provided by the above embodiment, and the other technical features in the interactive control device are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0243] In addition, to achieve the above objectives, the present application also proposes an intelligent terminal, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the interactive control method described above.

[0244] Reference below Figure 7 , which shows a schematic diagram of the structure of an intelligent terminal suitable for implementing the embodiments of the present application. The intelligent terminal in the embodiments of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 7 The smart terminal shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0245] like Figure 7As shown, the intelligent terminal may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the intelligent terminal are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the intelligent terminal to communicate with other devices wirelessly or by wire to exchange data. Although the figure shows an intelligent terminal with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or have alternatively.

[0246] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0247] The intelligent terminal provided by this application adopts the interactive control method in the above embodiment, which can significantly improve the recognition accuracy of fraud or harassment risks. At the same time, combined with personalized early warning measures, it can achieve intelligent early warning and personalized defense, improve user experience and security, and thus provide users with a safe, efficient and personalized communication environment. Compared with the prior art, the beneficial effects of the intelligent terminal provided by this application are the same as the beneficial effects of the interactive control method provided in the above embodiment, and the other technical features in the intelligent terminal are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0248] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0249] In addition, to achieve the above-mentioned purpose, the present application also proposes a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the interactive control method described above are implemented.

[0250] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0251] The computer-readable storage medium may be included in the smart terminal, or may exist independently without being installed in the smart terminal.

[0252] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the smart terminal, the smart terminal: performs a risk assessment on the interaction scenario based on the interaction information and through a preset risk assessment model; when it is determined that there is a risk in the interaction scenario, the smart terminal executes corresponding defense measures according to the risk assessment results.

[0253] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0254] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0255] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0256] The readable storage medium provided by the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned interactive control method, which can significantly improve the recognition accuracy of fraud or harassment risks, and at the same time, combined with personalized early warning measures, can achieve intelligent early warning and personalized defense, improve user experience and security, and thus provide users with a safe, efficient, and personalized communication environment. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as the beneficial effects of the interactive control method provided by the above-mentioned embodiment, which will not be repeated here.

[0257] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the interactive control method described above are implemented.

[0258] The computer program product provided by this application can significantly improve the recognition accuracy of fraud or harassment risks, and at the same time, combined with personalized early warning measures, can achieve intelligent early warning and personalized defense, improve user experience and security, and thus provide users with a safe, efficient, and personalized communication environment. Compared with the prior art, the beneficial effects of the computer program product provided by this application are the same as the beneficial effects of the interactive control method provided by the above embodiment, which will not be repeated here.

[0259] One or more technical solutions proposed in this application have at least the following technical effects: by collecting interactive information in interactive scenarios; based on the interactive information and through a preset risk assessment model, the interactive scenario is risk assessed; when it is determined that the interactive scenario is risky, the intelligent terminal executes corresponding defense measures according to the risk assessment results. This application uses deep neural network models and other methods to conduct risk assessments on interactive scenarios, which can significantly improve the accuracy of identifying fraud or harassment risks. At the same time, combined with personalized early warning measures, it can achieve intelligent early warning and personalized defense, improve user experience and security, and thus provide users with a safe, efficient, and personalized communication environment. This application has broad application prospects and important social value in the field of telecommunications fraud and harassment call defense.

[0260] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. An interactive control method, characterized in that: Applied to a smart terminal, the method comprises: Collect interaction information in interactive scenarios; Based on the interaction information, and using a preset risk assessment model, a risk assessment is performed on the interaction scenario; When it is determined that the interaction scenario has risks, the smart terminal executes corresponding defense measures according to the risk assessment result.

2. The method according to claim 1, characterized in that The step of performing risk assessment on the interaction scenario based on the interaction information and through a preset risk assessment model includes: Extracting features from the interaction information to obtain key feature information, where the key feature information includes: one or more of voice features, text features, and picture features; Converting the key feature information into a key feature vector; The key feature vector is input into a preset risk assessment model, the key feature vector is analyzed and identified through the risk assessment model, and the fraud or harassment risk of the interaction scenario is assessed based on the analysis and identification results.

3. The method according to claim 2, characterized in that The step of analyzing and identifying the key feature vector by using the risk assessment model, and performing fraud or harassment risk assessment on the interaction scenario according to the analysis and identification result comprises: By using the risk assessment model, combined with the pre-built fraud or harassment feature library and anti-fraud or anti-harassment feature library, the key feature vector is analyzed and identified to obtain an initial analysis and identification result; The initial analysis and identification results are cross-validated in combination with the pre-built cloud database and local database, and the fraud or harassment risk of the interaction scenario is assessed based on the cross-validation results.

4. The method according to claim 3, characterized in that Before the step of combining the pre-built cloud database and the local database to cross-validate the initial analysis and recognition results, the step further includes: When constructing the cloud database, desensitization processing is performed on the data in the cloud database.

5. The method according to claim 1, characterized in that The training steps of the risk assessment model include: Building a fraud or harassment feature library, an anti-fraud or anti-harassment feature library based on historical interaction data and / or constructed interaction data; Using the fraud or harassment feature library and the anti-fraud or anti-harassment feature library, model training is performed based on a deep neural network framework to obtain an initial deep neural network model; Based on a preset tuning strategy, the parameters of the initial deep neural network model are tuned to obtain a preset risk assessment model.

6. The method according to claim 5, characterized in that The fraud or harassment feature library includes: one or more features of impersonation, creation of emergency or tempting situations, professional rhetoric and techniques, and request for sensitive information or transfer of funds; the anti-fraud or anti-harassment feature library includes: one or more features of social network, timbre, accent, language habits, and social object portraits.

7. The method according to any one of claims 1 to 6, characterized in that In the case where it is determined that the interaction scenario has a risk, the step of the smart terminal executing corresponding defense measures according to the risk assessment result includes: In the case where it is determined that there is a risk of fraud or harassment in the interaction scenario, the smart terminal calls a pre-defined defense measure corresponding to the user in the current interaction scenario according to the risk assessment result; Execute the defense measures corresponding to the user in the current interaction scenario, or execute corresponding defense measures based on the priority of the defense measures in the current interaction scenario.

8. The method according to claim 7, characterized in that The interaction scenarios include: one or more of voice call scenarios, text chat scenarios, and video interaction scenarios; the defense measures include: interrupting the current interaction, voice or text message reply, adding the other party's number to a blacklist or reporting; the priorities include: reminder, warning, intervention, and interruption.

9. The method according to any one of claims 1 to 6, characterized in that In the case where it is determined that the interaction scenario has a risk, before the step of the smart terminal executing corresponding defense measures according to the risk assessment result, the step further includes: Develop corresponding defense measures for users through user behavior pattern mining models.

10. The method according to claim 9, characterized in that The step of formulating corresponding defense measures for users through the user behavior pattern mining model includes: Through the user behavior pattern mining model, the user behavior characteristics in the corresponding interaction scenarios in the historical interaction data are mined and analyzed to obtain the user interaction behavior pattern; Based on the user behavior characteristics and user interaction behavior patterns, construct a user behavior feature library; Corresponding defense measures are formulated for the user based on the user behavior feature library.

11. The method according to claim 10, characterized in that The step of mining and analyzing the user behavior characteristics in the corresponding interaction scenarios in the historical interaction data by using the user behavior pattern mining model includes: By using the user behavior pattern mining model, mining the interaction behavior features in the corresponding interaction scenarios in the historical interaction data, wherein the interaction behavior features include: one or more of voice features, text features, and picture features; The interactive behavior characteristics are analyzed from multiple dimensions to obtain a user interactive behavior pattern, wherein the multiple dimensions include: one or more of interaction duration, interaction frequency, and temporal communication relationship.

12. The method according to claim 10, characterized in that The method further comprises: Collect new interaction data, and based on the new interaction data, update the parameters of the risk assessment model to obtain an updated risk assessment model; and / or update the user behavior feature library based on the new interaction data and user behavior features.

13. The method according to any one of claims 1 to 6, characterized in that The intelligent terminal is equipped with an anti-fraud decision-making agent, and the step of performing risk assessment on the interaction scenario based on the interaction information and through a preset risk assessment model includes: The anti-fraud decision-making agent calls a preset risk assessment model based on the interaction information to perform risk assessment on the interaction scenario, and outputs defense measures if it is assessed that there is a risk in the interaction scenario.

14. An intelligent terminal, characterized in that: The intelligent terminal comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the interactive control method according to any one of claims 1 to 13.

15. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, which, when executed by a processor, implements the steps of the interactive control method according to any one of claims 1 to 13.

16. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the interactive control method according to any one of claims 1 to 13 are implemented.

Images