Telephone data authenticity prediction method and device, equipment and storage medium

Abstract

The invention provides a telephone data authenticity prediction method and device, equipment and a storage medium, and relates to the technical field of machine learning. Generating a training data set and a test data set according to the telephone data set; obtaining an original authenticity prediction model; taking the training data and the business source as input, taking the real label as output, and carrying out training processing on the original authenticity prediction model to obtain a trained authenticity prediction model; performing iterative training on the trained authenticity prediction model by adopting an expectation maximization algorithm according to the test data set until a preset iteration requirement is met, and obtaining a target prediction model; wherein the target prediction model is used for outputting a target prediction result according to the input to-be-predicted telephone data and a service source corresponding to the to-be-predicted telephone data.

Description

Technical Field

[0001] The present application relates to the field of machine learning technology, and in particular to a method, apparatus, device and storage medium for predicting the authenticity of telephone data. Background Art

[0002] In scenarios such as insurance, finance, and telecommunications, telephone communication is a vital part of business operations. A large amount of telephone data is stored in various business systems. Accurately determining the authenticity of telephone data is crucial to ensuring the normal operation of business and preventing risks.

[0003] In related technologies, the authenticity of phone data is mainly judged by manual experience. Specifically, the authenticity probability of phone data and its judgment threshold are manually set in the insurance process to score the authenticity of phone data.

[0004] However, the accuracy of the authenticity prediction of telephone data in the existing technology is low. Summary of the Invention

[0005] The present application provides a method, apparatus, device and storage medium for predicting the authenticity of telephone data, thereby solving the problem of low accuracy of authenticity prediction of telephone data in the prior art.

[0006] In a first aspect, the present application provides a method for predicting the authenticity of telephone data, comprising:

[0007] Acquire a telephone data set; wherein the telephone data set includes a plurality of telephone data and a business source corresponding to the telephone data;

[0008] Generate a training data set and a test data set based on the telephone data set; wherein the training data set includes a plurality of training data, a business source corresponding to the training data, and a true label corresponding to the training data;

[0009] Obtain the original authenticity prediction model;

[0010] Taking the training data and the business source as input and the true label as output, the original authenticity prediction model is trained to obtain a trained authenticity prediction model;

[0011] According to the test data set, the trained authenticity prediction model is iteratively trained using the maximum expectation algorithm until the preset iteration requirements are met, thereby obtaining a target prediction model; wherein, the target prediction model is used to output a target prediction result based on the input telephone data to be predicted and the business source corresponding to the telephone data to be predicted.

[0012] The present application provides a method for accurately determining the authenticity of telephone data. First, an authenticity prediction model is trained using a telephone data set containing business sources, which can learn complex feature patterns from telephone data from multiple business sources. Then, the authenticity prediction model is iteratively optimized using an expectation-maximization (EM) algorithm, which ensures the accuracy of the authenticity prediction model in the prediction and recognition process, and enhances the adaptability of the authenticity prediction model. It can still maintain high performance in the face of constantly changing data patterns, thereby improving the accuracy of telephone data authenticity prediction.

[0013] Optionally, the iterative training of the trained authenticity prediction model using a maximum expectation algorithm according to the test data set until a preset iteration requirement is met to obtain a target prediction model includes:

[0014] In the E-step calculation process of the maximum expectation algorithm, the trained authenticity prediction model is used to perform forward propagation processing on the test data set to obtain a test authenticity prediction probability;

[0015] Determining a test set score based on the predicted probability of test authenticity;

[0016] In the M-step calculation process of the maximum expectation algorithm, adjusting the model parameters and the judgment threshold of the trained authenticity prediction model according to the test set score;

[0017] Continue to train the adjusted authenticity prediction model until the preset iteration requirements are met to obtain the target prediction model.

[0018] Here, this application adopts the EM algorithm to realize the iterative optimization of the authenticity prediction model, updates the probability score through the E step, and synchronously adjusts the parameters and thresholds through the M step. The joint optimization of the E step and the M step can realize the parameter optimization of the authenticity prediction model in the process of alternating iterations, further improving the accuracy of determining the authenticity of telephone data.

[0019] Optionally, adjusting the model parameters and decision threshold of the trained authenticity prediction model according to the test set score includes:

[0020] According to the test set score, adjusting the model parameters of the trained authenticity prediction model using a cross entropy loss function, an adaptive moment estimation optimizer, and a cosine annealing strategy;

[0021] The model parameters of the trained authenticity prediction model are fixed, and the determination threshold of the trained authenticity prediction model is adjusted by a preset determination threshold loss function.

[0022] Among them, this application adopts the cross-entropy loss function, Adaptive Moment Estimation optimizer (AdamW) optimizer and cosine annealing strategy to adjust the model parameters of the authenticity prediction model. The cross-entropy loss function can accurately predict the gap between the probability distribution and the true label. Through the adjustment of the cross-entropy loss function, the prediction result of the authenticity prediction model can be made as close to the true label as possible. The participation of the AdamW optimizer is used to optimize the weight attenuation coefficient and the learning rate, which can reduce the overfitting of manually labeled data and improve the optimization efficiency of the model. The cosine annealing strategy provides a smooth and effective way to dynamically adjust the learning rate by utilizing the characteristics of the cosine function. Based on the adjustment of the above model parameters, the model parameters are fixed and the judgment threshold is adjusted, which can efficiently realize the optimization training of the authenticity prediction model, further improving the accuracy and efficiency of the authenticity prediction of telephone data.

[0023] Optionally, before adjusting the model parameters and decision threshold of the trained authenticity prediction model according to the test set score, the method further includes:

[0024] Obtain a verification data set and determine the verification set score;

[0025] Accordingly, adjusting the model parameters and decision thresholds of the trained authenticity prediction model according to the test set score includes:

[0026] According to the test set score and the verification set score, the model parameters and the judgment threshold of the trained authenticity prediction model are adjusted.

[0027] Here, in order to evaluate the performance of the model, this application constructs a verification dataset, which can provide a basis for the calculation of the loss function.

[0028] Optionally, the preset iteration requirement includes:

[0029] If the difference between the decision thresholds obtained from two consecutive trainings is less than the preset difference threshold, the iteration is stopped; and\or,

[0030] If the total number of training times is greater than the preset training threshold, the iteration is stopped.

[0031] Here, the present application sets flexible iteration conditions. If the difference between the judgment thresholds obtained in two consecutive iterations is less than a preset difference threshold, the iteration is stopped. The judgment threshold obtained at this time is the optimal judgment threshold. Alternatively, the present application also sets an early stopping mechanism. If the number of iterations is too many, the iteration is stopped. This improves the efficiency of model training while ensuring the accuracy of model training.

[0032] Optionally, generating a training dataset and a test dataset based on the telephone dataset includes:

[0033] Performing standardization and cleaning on the telephone data to obtain first preprocessed data;

[0034] Performing structured processing on the first preprocessed data to obtain a preprocessed data set;

[0035] Splitting the preprocessed data set according to a preset ratio to obtain a test data set and a training data sample; wherein the training data sample includes a plurality of training data and business sources corresponding to the training data;

[0036] Obtaining true labels corresponding to the training data;

[0037] A training data set is generated according to the true label and the training data samples.

[0038] Here, before predicting the authenticity of telephone data, this application first performs standardized cleaning and structured processing of the data, which can effectively eliminate useless, repeated or erroneous information, ensure the quality and accuracy of the data set, and then obtain the real labels corresponding to the training data, and then generate accurate and easy-to-process test data sets and training data sets, providing accurate data for model training, improving the training accuracy and training efficiency of the authenticity prediction model, and further improving the accuracy of determining the authenticity of telephone data.

[0039] Optionally, after iteratively training the trained authenticity prediction model using a maximum expectation algorithm according to the test data set until a preset iteration requirement is met and a target prediction model is obtained, the method further includes:

[0040] Acquire the telephone data to be predicted and the service source corresponding to the telephone data to be predicted;

[0041] A target prediction result is determined according to the target prediction model, the telephone data to be predicted, and the business source corresponding to the telephone data to be predicted.

[0042] Among them, this application can use a target prediction model to predict the authenticity of telephone data based on the telephone data to be predicted and the business source corresponding to the telephone data to be predicted, output accurate authenticity prediction results, and improve the accuracy of determining the authenticity of telephone data.

[0043] In a second aspect, the present application provides a device for predicting the authenticity of telephone data, comprising:

[0044] A first acquisition module is configured to acquire a telephone data set, wherein the telephone data set includes a plurality of telephone data and a service source corresponding to the telephone data;

[0045] A generation module, configured to generate a training data set and a test data set based on the telephone data set; wherein the training data set includes a plurality of training data, a service source corresponding to the training data, and a true label corresponding to the training data;

[0046] The second acquisition module is used to obtain the original authenticity prediction model;

[0047] a training module, configured to take the training data and the business source as input and the true label as output, and perform training processing on the original authenticity prediction model to obtain a trained authenticity prediction model;

[0048] An iteration module is used to iteratively train the trained authenticity prediction model based on the test data set using a maximum expectation algorithm until a preset iteration requirement is met, thereby obtaining a target prediction model; wherein, the target prediction model is used to output a target prediction result based on the input telephone data to be predicted and the business source corresponding to the telephone data to be predicted.

[0049] Optionally, the iteration module specifically includes:

[0050] A first iterative submodule is configured to perform forward propagation processing on the test data set using the trained authenticity prediction model during the E-step calculation process of the maximum expectation algorithm to obtain a test authenticity prediction probability;

[0051] A second iterative submodule is configured to determine a test set score based on the test authenticity prediction probability;

[0052] A third iterative submodule is configured to adjust the model parameters and the judgment threshold of the trained authenticity prediction model according to the test set score during the M-step calculation process of the maximum expectation algorithm;

[0053] The fourth iterative submodule is used to continue training the adjusted authenticity prediction model until the preset iteration requirements are met to obtain the target prediction model.

[0054] Optionally, the third iterator module is specifically configured to:

[0055] According to the test set score, adjusting the model parameters of the trained authenticity prediction model using a cross entropy loss function, an adaptive moment estimation optimizer, and a cosine annealing strategy;

[0056] The model parameters of the trained authenticity prediction model are fixed, and the determination threshold of the trained authenticity prediction model is adjusted by a preset determination threshold loss function.

[0057] Optionally, before the third iterative submodule is used to adjust the model parameters and decision threshold of the trained authenticity prediction model according to the test set score, it also includes:

[0058] The fifth iterative submodule is used to obtain the verification data set and determine the verification set score;

[0059] Accordingly, the third iterator module is specifically configured to:

[0060] According to the test set score and the verification set score, the model parameters and the judgment threshold of the trained authenticity prediction model are adjusted.

[0061] Optionally, the preset iteration requirement includes:

[0062] If the difference between the decision thresholds obtained from two consecutive trainings is less than the preset difference threshold, the iteration is stopped; and\or,

[0063] If the total number of training times is greater than the preset training threshold, the iteration is stopped.

[0064] Optionally, the generating module is specifically configured to:

[0065] Performing standardization and cleaning on the telephone data to obtain first preprocessed data;

[0066] Performing structured processing on the first preprocessed data to obtain a preprocessed data set;

[0067] Splitting the preprocessed data set according to a preset ratio to obtain a test data set and a training data sample; wherein the training data sample includes a plurality of training data and business sources corresponding to the training data;

[0068] Obtaining true labels corresponding to the training data;

[0069] A training data set is generated according to the true label and the training data samples.

[0070] Optionally, after the iteration module is used to iteratively train the trained authenticity prediction model using a maximum expectation algorithm based on the test data set until a preset iteration requirement is met and a target prediction model is obtained, the apparatus further includes a result determination module for:

[0071] Acquire the telephone data to be predicted and the service source corresponding to the telephone data to be predicted;

[0072] A target prediction result is determined according to the target prediction model, the telephone data to be predicted, and the business source corresponding to the telephone data to be predicted.

[0073] In a third aspect, the present application provides a device for predicting authenticity of telephone data, comprising: at least one processor and a memory;

[0074] The memory stores computer-executable instructions;

[0075] The at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor performs the authenticity prediction method of telephone data as described in the first aspect and various possible designs of the first aspect.

[0076] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions. When a processor executes the computer-executable instructions, the authenticity prediction method of telephone data as described in the first aspect and various possible designs of the first aspect is implemented.

[0077] The present application provides a method, apparatus, device and storage medium for predicting the authenticity of telephone data. Before predicting the authenticity of telephone data, the method first performs standardized cleaning and structured processing of the data, which can effectively eliminate useless, repeated or erroneous information, ensure the quality and accuracy of the data set, and then obtain the true labels corresponding to the training data, thereby generating accurate and easy-to-process test data sets and training data sets, providing accurate data for model training, improving the training accuracy and training efficiency of the authenticity prediction model, and further improving the accuracy of determining the authenticity of telephone data. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] 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.

[0079] Figure 1 A schematic diagram of the architecture of a system for predicting the authenticity of telephone data provided in an embodiment of the present application;

[0080] Figure 2 A flowchart of a method for predicting the authenticity of telephone data provided in an embodiment of the present application;

[0081] Figure 3 A schematic diagram of the training process of a authenticity prediction model provided in an embodiment of the present application;

[0082] Figure 4 A schematic diagram of the structure of a device for predicting the authenticity of telephone data provided in an embodiment of the present application;

[0083] Figure 5 A schematic diagram of the structure of a device for predicting the authenticity of telephone data provided in an embodiment of the present application.

[0084] The above drawings illustrate specific embodiments of the present application, which will be described in more detail below. These drawings and the textual description are not intended to limit the scope of the present application in any way, but rather to illustrate the concepts of the present application to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0085] Exemplary embodiments are described in detail herein, with examples illustrated in the accompanying drawings. When the following description refers to the drawings, identical numerals in different figures represent identical or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with the present application. Rather, they are merely examples of apparatuses and methods consistent with certain aspects of the present application.

[0086] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0087] First, let’s introduce the nouns that will appear in the examples of this application:

[0088] Insurance business processes include underwriting, payment, reporting, inspection, damage assessment, claims settlement, and follow-up visits. Each phase requires recording customer contact information. For example, during the underwriting phase, the customer proactively provides information, generating an initial phone call record. Here, customer contact information is an optional form of phone data.

[0089] Telephone collection: Throughout the insurance business process (underwriting, claims settlement, follow-up visits, etc.), customer information (including name, ID number) and telephone number must undergo standardized cleaning (such as number format verification and information deduplication) and integrity verification. These data are then integrated into a structured data set using a unique customer identifier (such as a customer code). The data includes basic customer information, contact information (telephone number), business process identifier, and collection timestamp, and are ultimately stored in a relational database.

[0090] Customer Call Scoring: In the insurance business, call authenticity scoring rules are manually and statically set with a probability of authenticity at each stage (e.g., 80% for underwriting, 90% for claims). These probabilities are based on historical business experience. For example, the claims stage, due to the need to verify the authenticity of an accident, has a default probability higher than the underwriting stage. The threshold for determining authenticity is manually preset (e.g., 60 points). A call is considered authentic if it scores ≥ 60 points. However, the threshold adjustment cycle is fixed and cannot respond to sudden data fluctuations in real time.

[0091] Deep neural networks are advanced models in the field of machine learning. They employ multiple hidden layers and nonlinear transformations to capture complex relationships and patterns in data. The input layer of a DNN receives raw data, while the hidden layers gradually abstract and transform it. Ultimately, the output layer generates predictions or classifications.

[0092] Early Termination: Early termination, also known as early stopping, is an effective model regularization method designed to prevent overfitting during deep neural network training. This strategy monitors performance on the validation set and stops training when performance begins to decline. This approach prevents the model from overfitting the training data while maintaining good generalization to unseen data.

[0093] Adaptive Weight Update Algorithm: The AdamW optimizer is an improvement on the Adam optimizer. It implements adaptive updates to the weights of deep neural networks by applying L2 regularization to the weight decay term. AdamW combines momentum with an adaptive learning rate adjustment mechanism to efficiently update network parameters while reducing the risk of overfitting.

[0094] In scenarios such as insurance, finance, and telecommunications, telephone communication is a vital part of business operations. A large amount of telephone data is stored in various business systems. Accurately determining the authenticity of telephone data is crucial to ensuring the normal operation of business and preventing risks.

[0095] Currently, there are two main technical solutions for calculating the authenticity of phone calls.

[0096] Solution 1: Experience-Based Method. This approach manually sets the probability of authenticity of phone calls and their thresholds for determining authenticity during the insurance process. The core issue with this approach is its reliance on individual subjective judgment and past experience, resulting in inaccurate scoring and low customer identification accuracy. Furthermore, the lack of an objective standard to measure the effectiveness of these manually set parameters further limits the reliability and applicability of this approach.

[0097] Solution 2: Based on the Bayesian formula algorithm. This approach iteratively optimizes relevant probability values ​​through underlying algorithms, but encounters several key obstacles in practice. First, the Bayesian approach requires explicit assumptions about the probability distribution of the data, making the model inefficient when dealing with high-dimensional or nonlinear data. Second, the feature engineering process relies on expert knowledge to manually design rules to extract business-related features, making it impossible to automatically discover complex associations hidden in the data. Finally, within the EM algorithm framework, probability updates are based solely on iterative calculations of statistics, failing to effectively capture changes in the underlying semantic features of the data, limiting the model's ability to adapt to dynamic environments.

[0098] Existing technical solutions are primarily based on manual experience and Bayesian algorithms. The main drawback of these solutions is their heavy reliance on subjective judgment and accumulated experience, lacking objective criteria, which leads to inaccurate scoring and low customer identification accuracy. While Bayesian algorithms attempt to address the challenges inherent in manual experience through algorithmic optimization, they still suffer from significant deficiencies in handling complex data structures, automated feature extraction, and dynamic adaptability. Consequently, existing technologies for predicting the authenticity of phone data have low accuracy.

[0099] In order to solve the above problems, the embodiments of the present application provide a method, apparatus, device and medium for predicting the authenticity of telephone data. The method trains an authenticity prediction model by using a telephone data set containing business sources, and can learn complex feature patterns from telephone data from multiple business sources. Afterwards, the authenticity prediction model is iteratively optimized using the EM algorithm, thereby obtaining an authenticity prediction model that can accurately determine the authenticity of the telephone data.

[0100] Furthermore, one possible implementation incorporates a DNN and an improved EM framework to achieve precise prediction and dynamic optimization of the phone authenticity probability. This approach significantly improves scoring accuracy and the model's dynamic adaptability, making it suitable for customer authentication scenarios across multiple sectors, including insurance, finance, and telecommunications. This integrated approach not only enhances the accuracy of phone authenticity assessments but also ensures the model's reliability and robustness across a wide range of application scenarios, making it a powerful tool for improving the efficiency and accuracy of customer authentication.

[0101] Optional, Figure 1 This is a schematic diagram of the architecture of a phone data authenticity prediction system provided in an embodiment of the present application. Figure 1 In the above architecture, the above architecture includes at least one of a data acquisition device 101, a processing device 102 and a display device 103.

[0102] It is understood that the structure illustrated in the embodiments of this application does not constitute a specific limitation on the architecture of the phone data authenticity prediction system. In other feasible implementations of this application, the above architecture may include more or fewer components than shown, or may combine or split certain components, or have different component arrangements. The specific configuration can be determined based on the actual application scenario and is not limited here. Figure 1 The components shown can be implemented in hardware, software, or a combination of software and hardware.

[0103] In a specific implementation process, the data acquisition device 101 may include an input / output interface and may also include a communication interface. The data acquisition device 101 may be connected to the processing device via the input / output interface or the communication interface.

[0104] The processing device 102 trains the authenticity prediction model by using a telephone data set containing business sources, and can learn complex feature patterns from telephone data from multiple business sources. Afterwards, the authenticity prediction model is iteratively optimized using the EM algorithm to obtain an authenticity prediction model that can accurately determine the authenticity of the telephone data.

[0105] The display device 103 may also be a touch screen display or a screen of a terminal device, which is used to receive user instructions while displaying the above content to achieve interaction with the user.

[0106] It should be understood that the above-mentioned processing device can be implemented by a processor reading instructions in a memory and executing the instructions, or it can be implemented by a chip circuit.

[0107] In addition, the network architecture and business scenarios described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Ordinary technicians in this field can know that with the evolution of network architecture and the emergence of new business scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0108] The technical solution of this application is described in detail below with reference to specific embodiments:

[0109] Optionally, Figure 2 A flowchart of a method for predicting the authenticity of telephone data provided in an embodiment of the present application. The execution subject of the embodiment of the present application may be Figure 1 The specific execution subject of the processing device 102 can be determined according to the actual application scenario. Figure 2 As shown, the method includes the following steps:

[0110] S201: Acquire a telephone data set.

[0111] The telephone data set includes multiple telephone data and business sources corresponding to the telephone data.

[0112] Optionally, the telephone data here includes telephone number (also called telephone number), mobile phone number or other contact information.

[0113] Optionally, when the application scenario of the embodiment of the present application is an insurance business process, the business source is an application link in the insurance business process, such as underwriting, claims settlement, and return visits.

[0114] S202: Generate a training dataset and a test dataset based on the phone dataset.

[0115] The training data set includes multiple training data, business sources corresponding to the training data, and real labels corresponding to the training data.

[0116] Optionally, generating a training dataset and a test dataset based on the phone dataset includes:

[0117] The telephone data is standardized and cleaned to obtain first preprocessed data; the first preprocessed data is structured to obtain a preprocessed data set; the preprocessed data set is split according to a preset ratio to obtain a test data set and a training data sample; wherein the test data sample includes multiple test data and business sources corresponding to the test data; the true label corresponding to the training data is obtained; and a training data set is generated based on the true label and the training data sample.

[0118] Optionally, the real label here can be input by the user through an input / output device, a display device, etc., or sent by a terminal device of the receiving user.

[0119] It is understandable that the preset ratio here can be determined according to actual conditions, and the embodiments of the present application do not impose any specific restrictions on this.

[0120] Optionally, the training data set includes multiple training data and business sources corresponding to the training data.

[0121] Both the training data and the test data here belong to phone data.

[0122] In a possible implementation, S201-S202 are implemented as follows:

[0123] Collect existing phone records containing data sources and set scoring criteria based on the application of each record in the insurance business process. At the same time, clean the collected data and remove duplicate, erroneous or invalid information to ensure the quality and accuracy of the data set. After the above steps, a valid phone data set with a total number of n is finally formed, where n is any positive integer. This data set can be defined as a specific structured phone data set v = {v1, v2, ..., v n}, v i ∈R n . Among them, each sample data v i Contains data service source identification. At the same time, each phone has a corresponding data service source identification. For a total number of data source identifications of m, where m is any positive integer, the structured phone data set can be defined as x i ={x i1 , x i2 ,...,x im}, x i ∈R n+m , represents the sample x i The structured telephone data is the pre-processed data set in the embodiment of the present application.

[0124] Here, before predicting the authenticity of telephone data, the embodiment of the present application first performs standardized cleaning and structured processing of the data, which can effectively eliminate useless, repeated or erroneous information, ensure the quality and accuracy of the data set, and then obtain the real labels corresponding to the training data, and then generate accurate and easy-to-process test data sets and training data sets, providing accurate data for model training, improving the training accuracy and training efficiency of the authenticity prediction model, and further improving the accuracy of determining the authenticity of telephone data.

[0125] S203: Obtain the original authenticity prediction model.

[0126] Optionally, the authenticity prediction model is a DNN model.

[0127] Optionally, an original authenticity prediction model is pre-established, wherein the authenticity prediction model includes an input layer, a hidden layer, and an output layer.

[0128] Input layer: The input layer uses one-hot encoding and retains the data x i The original distribution characteristics of the data are obtained to avoid the loss of individual information, and discrete categories are mapped into high-dimensional sparse vectors. The sparsity is used to enhance the model's ability to distinguish the source of industry data in a refined manner. The sparse matrix compression storage and parallel computing optimization are combined to achieve efficient feature expression and computing performance.

[0129] Hidden layer: After the high-dimensional sparse vector of the input layer is linearly projected, most neurons are in the negative range. Through the "unilateral inhibition" characteristic of the Rectified Linear Unit (ReLU), invalid signals such as redundant coding bits are automatically filtered out, and only strong related features are transmitted to reduce model noise interference. For the input vector x i After nonlinear transformation and ReLU activation function, dynamic screening of feature space and high-order interaction modeling are achieved. The specific formula is as follows:

[0130] h1=ReLU(W1·x1+b1)

[0131] In the hidden layer of layer l, dynamic screening of feature space and high-order interaction modeling are implemented as follows:

[0132]

[0133] Among them, h1 is the output of the first hidden layer after nonlinear transformation (through ReLU activation function), is the output of the i-th neuron in the l-th hidden layer after nonlinear transformation (through ReLU activation function). W1 is the weight matrix of the first layer, W il is the weight matrix of the lth layer, is the bias value of the lth layer. ReLU performs threshold truncation on the input, retaining only positive outputs and forcing negative values ​​to zero.

[0134] Output layer: For phone rating, it can be classified as a binary task, and the output vector of the hidden layer is mapped to [0, 1]. After extracting high-dimensional features through DNN, the activation vector of the last hidden layer needs to be Mapping to normalized scoring probability. The formal definition is:

[0135]

[0136] Among them, P(Scrore>S c |x i ) means that given an input vector x i Under the condition that the score is greater than the judgment threshold S c The probability of . W is the weight matrix, b is the bias value, and σ is the activation function sigmoid, that is,

[0137] S204: Taking the training data and the business source as input and the true label as output, the original authenticity prediction model is trained to obtain a trained authenticity prediction model.

[0138] S205: Based on the test data set, the trained authenticity prediction model is iteratively trained using the maximum expectation algorithm until the preset iteration requirements are met, thereby obtaining a target prediction model.

[0139] Optionally, the method of iteratively training the trained authenticity prediction model using a maximum expectation algorithm according to the test data set until a preset iteration requirement is met to obtain a target prediction model includes:

[0140] During the E-step calculation process of the maximum expectation algorithm, the trained authenticity prediction model is used to perform forward propagation processing on the test data set to obtain the test authenticity prediction probability; based on the test authenticity prediction probability, the test set score is determined; during the M-step calculation process of the maximum expectation algorithm, the model parameters and judgment threshold of the trained authenticity prediction model are adjusted according to the test set score; the adjusted authenticity prediction model is continued to be trained until the preset iteration requirements are met, and the target prediction model is obtained.

[0141] In one possible implementation, in the E-step of the EM algorithm, the probability model of the current iteration is used to calculate the authenticity probability of the phone call of each data source, and each phone record in the data set is re-scored, as follows.

[0142] Step E - Score Calculation: In the insurance business scenario, according to the above probability model (authenticity prediction model), the authenticity probability of the call in each insurance business link can be calculated. i Re-scoring is performed to update the call assessment score for the next iteration of new probability statistics.

[0143] V i =P(Scrore>S c |x i )*100

[0144] Optionally, the test data set here can be the test data set obtained through the telephone data set mentioned in the above embodiment, or all the data after preprocessing of the above telephone data set, or it can be a test data set sent by the receiving user. There is no specific limitation here. The i in different formulas in the embodiments of the present application can be the same or different.

[0145] Optionally, before adjusting the model parameters and judgment threshold of the trained authenticity prediction model according to the test set score, it also includes: obtaining a verification data set and determining the verification set score; accordingly, adjusting the model parameters and judgment threshold of the trained authenticity prediction model according to the test set score, including: adjusting the model parameters and judgment threshold of the trained authenticity prediction model according to the test set score and the verification set score.

[0146] At the same time, in order to evaluate the performance of the model, a manually annotated verification phone data set with a total number of o (o is an arbitrary positive integer) is constructed. The verification phone data set can be defined as R = {R1, R2, ..., R o}, for the verification data set R o Re-scoring is performed for subsequent calculation of loss function.

[0147] R o =P(Scrore>S c |x o )*100

[0148] M-step-maximize expected value: for the decision threshold S c To optimize, we need to find S that maximizes the expectation. c For example, after calculating the probability of each sample belonging to the positive class in the E step, the M step needs to adjust S c To maximize the likelihood of the overall data, or to minimize a certain loss function. Therefore, in step M of the embodiment of the present application, the AdamW optimizer is used to update the model parameters.

[0149] Optionally, based on the test set score, the model parameters and decision threshold of the trained authenticity prediction model are adjusted, including:

[0150] According to the test set score, the cross entropy loss function, adaptive moment estimation optimizer and cosine annealing strategy are used to adjust the model parameters of the trained authenticity prediction model; the model parameters of the trained authenticity prediction model are fixed, and the judgment threshold of the trained authenticity prediction model is adjusted through the preset judgment threshold loss function.

[0151] It is understandable that the preset judgment threshold loss function here can be determined according to actual conditions and is not specifically limited thereto.

[0152] Optionally, based on the test set score, the model parameters and decision threshold of the trained authenticity prediction model are adjusted, including:

[0153] According to the test set score, the model parameters of the trained authenticity prediction model are adjusted using the cross entropy loss function, adaptive moment estimation optimizer and cosine annealing strategy;

[0154] The model parameters of the trained authenticity prediction model are fixed, and the judgment threshold of the trained authenticity prediction model is adjusted through a preset judgment threshold loss function.

[0155] Among them, the embodiment of the present application adopts the cross-entropy loss function, the Adaptive Moment Estimation optimizer (AdamW) optimizer and the cosine annealing strategy to adjust the model parameters of the authenticity prediction model. The cross-entropy loss function can accurately predict the gap between the probability distribution and the true label. Through the adjustment of the cross-entropy loss function, the prediction result of the authenticity prediction model can be made as close to the true label as possible. The participation of the AdamW optimizer is used to optimize the weight attenuation coefficient and the learning rate, which can reduce the overfitting of manually labeled data and improve the optimization efficiency of the model. The cosine annealing strategy provides a smooth and effective way to dynamically adjust the learning rate by utilizing the characteristics of the cosine function. Based on the adjustment of the above-mentioned model parameters, the model parameters are fixed and the judgment threshold is adjusted, which can efficiently realize the optimization training of the authenticity prediction model, further improving the accuracy and efficiency of the authenticity prediction of telephone data.

[0156] Among them, the embodiment of the present application adopts the cross-entropy loss function, AdamW optimizer and cosine annealing strategy to adjust the model parameters of the authenticity prediction model. The cross-entropy loss function can accurately predict the gap between the probability distribution and the true label. Through the adjustment of the cross-entropy loss function, the prediction result of the authenticity prediction model can be made as close to the true label as possible. The participation of the AdamW optimizer is used to optimize the weight attenuation coefficient and the learning rate, which can reduce the overfitting of manually labeled data and improve the optimization efficiency of the model. The cosine annealing strategy provides a smooth and effective way to dynamically adjust the learning rate by utilizing the characteristics of the cosine function. Based on the adjustment of the above-mentioned model parameters, the model parameters are fixed and the judgment threshold is adjusted, which can efficiently realize the optimization training of the authenticity prediction model, further improving the accuracy and efficiency of the authenticity prediction of telephone data.

[0157] In one possible implementation, the specific parameter update of the M-step-maximized expected value includes the following strategy:

[0158] a. Loss function:

[0159] First, in classification tasks, the output of the model is usually a probability distribution, which represents the likelihood of each category. In order to find the optimal threshold for the evaluation metric in the M-step, the cross-entropy loss function is used to quantify the gap between the predicted probability distribution and the true label.

[0160] Through the loss function, the model parameters can be adjusted to make the model's prediction results as close to the true label as possible. At the same time, the embodiment of the application introduces L2 regularization to impose penalties on the weights, limit the complexity of the model, and prevent overfitting. The formula is:

[0161]

[0162] Among them, N is the total number of samples in the training set, K i is the true label of the i-th sample, and its value is [0, 1], where 1 represents true and 0 represents false. i represents the model's predicted probability for the i-th sample, λ is the regularization coefficient, and W is the hyperparameter that controls the regularization strength. l represents the weight matrix of layer l.

[0163] b. AdamW optimizer:

[0164] In order to optimize the weight decay coefficient λ and learning rate η t , the embodiment of the present application calculates the gradient of the loss function with respect to the parameters through back propagation, and adjusts the model parameters based on the gradient information to minimize the loss value. The formula is:

[0165]

[0166] Among them, θ t+1 and θ t Represent the parameter values ​​of the model at the t+1th and tth iterations, η t is the learning rate, which controls the step size of each parameter update. ∈ is a constant. t represents the number of update iterations, m t It represents an unbiased estimate of the gradient direction. It eliminates the error caused by the initial zero value through bias correction. It is approximately equal to the mean of the historical gradient, reflecting the overall trend of parameter update and guiding the direction of parameter update. t It represents an unbiased estimate of the gradient magnitude, which is used to measure the variance or volatility of the gradient and is used to adaptively adjust the learning rate.

[0167] c. Cosine annealing strategy:

[0168] The cosine annealing strategy provides a smooth and effective way to dynamically adjust the learning rate by utilizing the characteristics of the cosine function, which is used to adjust η in the AdamW optimizer. t The specific formula is:

[0169]

[0170] Among them, η max is the initial learning rate. In the early stages of training, a larger learning rate helps the model quickly approach the optimal solution. min is the minimum learning rate. In the later stages of training, a smaller learning rate helps the model fine-tune parameters and prevents training from stagnating due to a too low learning rate. T is the training cycle.

[0171] d. Optimize the judgment threshold (also called judgment score) S c :

[0172] In order to find the optimal threshold S c In this embodiment, the DNN parameter θ is fixed and the search [S min , S max ]. At the same time, the initial minimum loss minLoss=∞, the optimal decision For each candidate decision score, calculate the loss Loss(S i ), as follows:

[0173]

[0174] Among them, o represents the total number of samples in the verification data set, R o is the prediction score of the oth sample, K O represents the true label of the o-th sample, and |(·) represents the indicator function.

[0175] If Loss(S i) If <minLoss, update the optimal threshold and output the optimal decision threshold. The specific formula is as follows:

[0176]

[0177] Here, the embodiments of this application use the EM algorithm to iteratively optimize the authenticity prediction model. The probability scores are updated in the E step, and the parameters and thresholds are adjusted synchronously in the M step. The joint optimization of the E step and the M step can achieve the parameter optimization of the authenticity prediction model during the alternating iteration process, further improving the accuracy of determining the authenticity of telephone data.

[0178] Optionally, the preset iteration requirements include: if the difference between the decision thresholds obtained from two adjacent trainings is less than the preset difference threshold, stop the iteration; and / or, if the total number of trainings is greater than the preset training threshold, stop the iteration.

[0179] It can be understood that the preset difference threshold and the preset training threshold here can be determined according to the actual situation, and the embodiments of this application do not make specific limitations.

[0180] In a possible implementation manner, during the iteration process, if the difference between the decision scores obtained from two consecutive iterations is less than a preset threshold (preset difference threshold) ε, stop the iteration. At this time, the obtained decision score is the optimal decision score.

[0181] Therefore, the embodiments of this application record the optimal decision scores S of f iterations c as the set S = {S1, S2..., S f}, then when |S f - S f-1 | < ε, it is determined that the model converges and the iteration stops.

[0182] Optionally, when the iteration exceeds the preset number of iterations, it is considered that the decision score converges, and the decision score with the highest verification dataset accuracy rate is selected as the optimal decision score. At the same time, for the training process where the continuous iteration does not decrease, the embodiments of this application design an early stopping mechanism. The early stopping counter counts the number of cycles in which the loss does not decrease during the iteration process. When the calculation reaches the preset training threshold (for example, the calculation reaches 5), the iteration is terminated in advance.

[0183] Here, the embodiments of this application set flexible iteration conditions. If the difference between the decision thresholds obtained from two consecutive iterations is less than a preset difference threshold, stop the iteration. At this time, the obtained decision threshold is the optimal decision threshold. Or, the embodiments of this application also set an early stopping mechanism. If the number of iterations is too large, stop the iteration. On the basis of ensuring the training accuracy of the model, the training efficiency of the model is improved.

[0184] The target prediction model is used to output a target prediction result based on the input telephone data to be predicted and the business source corresponding to the telephone data to be predicted.

[0185] An embodiment of the present application provides a method for accurately determining the authenticity of telephone data. First, an authenticity prediction model is trained by using a telephone data set containing business sources, which can learn complex feature patterns from telephone data from multiple business sources. Then, the authenticity prediction model is iteratively optimized using the EM algorithm, which ensures the accuracy of the authenticity prediction model in the prediction and recognition process, and enhances the adaptability of the authenticity prediction model. It can still maintain high performance in the face of constantly changing data patterns, thereby improving the accuracy of telephone data authenticity prediction.

[0186] Optionally, after iteratively training the trained authenticity prediction model using a maximum expectation algorithm based on the test data set until a preset iteration requirement is met and a target prediction model is obtained, the method further includes:

[0187] Obtaining the telephone data to be predicted and the business source corresponding to the telephone data to be predicted;

[0188] The target prediction result is determined based on the target prediction model, the telephone data to be predicted, and the business source corresponding to the telephone data to be predicted.

[0189] Among them, the target prediction model can output a binary classification label based on the input phone data to be predicted and the business source corresponding to the phone data to be predicted. If the label is 1, it is determined that the phone data to be predicted is true. If the label is 0, it is determined that the phone data to be predicted is not true and is an error call.

[0190] Among them, the embodiment of the present application can use a target prediction model to predict the authenticity of telephone data based on the telephone data to be predicted and the business source corresponding to the telephone data to be predicted, output accurate authenticity prediction results, and improve the accuracy of determining the authenticity of telephone data.

[0191] In one possible implementation, Figure 3 As shown, the embodiment of the present application provides a schematic diagram of the training process of the authenticity prediction model, combined with Figure 3The core of the embodiment of the present application is as follows: (1) The introduction of DNN can automatically extract multi-dimensional features such as business links and call feedback information, thereby achieving accurate prediction of the probability of phone authenticity. Nonlinear modeling capabilities can capture complex relationships in the data, improving the accuracy and generalization ability of the scoring model; (2) The collaborative optimization strategy of the training framework adopts the cross entropy loss function combined with the L2 regularization term to balance the model fitting ability and generalization performance; the introduction of the AdamW optimizer improves the parameter update efficiency by decoupling the weight decay mechanism. At the same time, the cosine annealing learning rate scheduling strategy is adopted to dynamically adjust the training step size to jump out of the local optimal solution. Cooperating with the early stopping mechanism, the training time is shortened; (3) The design of the dynamic iteration framework, the E step and the M step are jointly optimized, the E step updates the probability score, and the M step synchronously adjusts the parameters and thresholds, and approaches the global optimal solution through alternating iterations.

[0192] By introducing a DNN, the model automatically extracts multi-dimensional features such as business processes and call feedback information, achieving accurate predictions of the probability of phone call authenticity. First, the DNN's nonlinear modeling capabilities capture complex relationships in the data, significantly improving the accuracy and generalization of the scoring model. The training framework utilizes a collaborative optimization strategy, combining a cross-entropy loss function with an L2 regularization term to balance the model's fit and generalization performance. The AdamW optimizer is introduced to improve parameter update efficiency by decoupling the weight decay mechanism. Furthermore, a cosine annealing learning rate scheduling strategy dynamically adjusts the training step size to escape local optima, and an early stopping mechanism is employed to shorten training time. Furthermore, a dynamic iteration framework is designed. During the E-step and M-step joint optimization process, the E-step updates the probability score, while the M-step simultaneously adjusts parameters and thresholds, approaching the global optimal solution through alternating iterations. This approach, combining insurance business data with the maximum expectation algorithm, achieves automatic probability updates, achieving higher accuracy than manual experience or Bayesian-based set values.

[0193] The present application proposes a joint DNN and EM optimization framework. This framework not only covers the design of the model architecture, but also includes the design of the loss function and a dynamic threshold adjustment mechanism to achieve efficient adaptation to different business scenarios. At the same time, considering the efficiency issues in practical applications, an optimization strategy is formulated to ensure the efficiency and stability of the model training process.

[0194] Compared to traditional Bayesian methods, the DNN+EM solution can more accurately capture data features, thereby providing more reliable prediction results. Secondly, the model has dynamic adaptability, automatically updating parameters and thresholds as the model trains, quickly adapting to new business scenarios without frequent manual intervention.

[0195] Figure 4 A schematic diagram of the structure of a device for predicting the authenticity of telephone data provided in an embodiment of the present application is shown in FIG. Figure 4As shown, the apparatus of the embodiment of the present application includes: a first acquisition module 401, a generation module 402, a second acquisition module 403, a training module 404, and an iteration module 405. The authenticity prediction device for phone data herein can be the aforementioned processing device, the processor itself, or a chip or integrated circuit that implements the functions of the processor. It should be noted that the division of first acquisition module 401, generation module 402, second acquisition module 403, training module 404, and iteration module 405 is merely a logical functional division; physically, the two can be integrated or independent.

[0196] in,

[0197] A first acquisition module is configured to acquire a telephone data set, wherein the telephone data set includes a plurality of telephone data and business sources corresponding to the telephone data;

[0198] A generation module is used to generate a training data set and a test data set based on the telephone data set; wherein the training data set includes a plurality of training data, a business source corresponding to the training data, and a true label corresponding to the training data;

[0199] The second acquisition module is used to obtain the original authenticity prediction model;

[0200] The training module is used to train the original authenticity prediction model using training data and business sources as input and true labels as output to obtain a trained authenticity prediction model;

[0201] The iteration module is used to iteratively train the trained authenticity prediction model based on the test data set using the maximum expectation algorithm until the preset iteration requirements are met, thereby obtaining the target prediction model; wherein the target prediction model is used to output the target prediction result based on the input telephone data to be predicted and the business source corresponding to the telephone data to be predicted.

[0202] Optionally, the iteration module specifically includes:

[0203] The first iterative submodule is used to perform forward propagation processing on the test data set using the trained authenticity prediction model during the E-step calculation process of the maximum expectation algorithm to obtain the test authenticity prediction probability;

[0204] The second iterative submodule is used to determine the test set score based on the test authenticity prediction probability;

[0205] The third iterative submodule is used to adjust the model parameters and decision thresholds of the trained authenticity prediction model according to the test set score during the M-step calculation process of the maximum expectation algorithm;

[0206] The fourth iterative submodule is used to continue training the adjusted authenticity prediction model until the preset iteration requirements are met to obtain the target prediction model.

[0207] Optionally, the third iterator module is specifically used to:

[0208] According to the test set score, the model parameters of the trained authenticity prediction model are adjusted using the cross entropy loss function, adaptive moment estimation optimizer and cosine annealing strategy;

[0209] The model parameters of the trained authenticity prediction model are fixed, and the judgment threshold of the trained authenticity prediction model is adjusted through a preset judgment threshold loss function.

[0210] Optionally, before the third iterative submodule is used to adjust the model parameters and decision threshold of the trained authenticity prediction model according to the test set score, the method further includes:

[0211] The fifth iterative submodule is used to obtain the verification data set and determine the verification set score;

[0212] Accordingly, the third iterator module is specifically used for:

[0213] According to the test set score and the verification set score, the model parameters and judgment threshold of the trained authenticity prediction model are adjusted.

[0214] Optionally, the preset iteration requirements include:

[0215] If the difference between the decision thresholds obtained from two consecutive trainings is less than the preset difference threshold, the iteration is stopped; and\or,

[0216] If the total number of training times is greater than the preset training threshold, the iteration is stopped.

[0217] Optionally, the generation module is specifically used to:

[0218] Standardizing and cleaning the telephone data to obtain first preprocessed data;

[0219] Performing structured processing on the first preprocessed data to obtain a preprocessed data set;

[0220] Splitting the preprocessed data set according to a preset ratio to obtain a test data set and a training data sample; wherein the training data sample includes multiple training data and the business sources corresponding to the training data;

[0221] Get the true labels corresponding to the training data;

[0222] Generate a training dataset based on the true labels and training data samples.

[0223] Optionally, after the iteration module is used to iteratively train the trained authenticity prediction model using the maximum expectation algorithm according to the test data set until the preset iteration requirements are met and the target prediction model is obtained, the above-mentioned device further includes a result determination module for:

[0224] Obtaining the telephone data to be predicted and the business source corresponding to the telephone data to be predicted;

[0225] The target prediction result is determined based on the target prediction model, the telephone data to be predicted, and the business source corresponding to the telephone data to be predicted.

[0226] refer to Figure 5 , which shows a schematic structural diagram of a device 500 for predicting the authenticity of telephone data suitable for implementing an embodiment of the present disclosure. The device 500 can be a terminal device or a server. The terminal device can include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, personal digital assistants (PDAs), tablet computers (Portable Android Devices, PADs), portable multimedia players (PMPs), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5 The authenticity prediction device for phone data shown is merely an example and should not limit the functions and scope of use of the embodiments of the present disclosure.

[0227] like Figure 5 As shown, the authenticity prediction device 500 for phone data may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes based on programs stored in a read-only memory (ROM) 502 or programs loaded from a storage device 508 into a random access memory (RAM) 503. RAM 503 also stores various programs and data required for the operation of the authenticity prediction device 500 for phone data. Processing device 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to bus 504.

[0228] Typically, the following devices may be connected to the I / O interface 505: an input device 506 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 507 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 508 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 509. The communication device 509 may allow the authenticity prediction device 500 of the phone data to communicate with other devices wirelessly or by wire to exchange data. Although Figure 5 The authenticity prediction apparatus 500 for phone data is shown with various devices, but it should be understood that it is not required to implement or possess all of the devices shown, and more or fewer devices may be implemented or possessed instead.

[0229] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network via the communication device 509, or installed from the storage device 508, or installed from the ROM 502. When the computer program is executed by the processing device 501, the above-mentioned functions defined in the method of the embodiment of the present disclosure are performed.

[0230] It should be noted that the computer-readable medium mentioned above in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, 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 the present disclosure, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, device, or component. In the present disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to wires, optical cables, RF (radio frequency), etc., or any suitable combination thereof.

[0231] The computer-readable medium may be included in the authenticity prediction device for telephone data, or may exist independently without being incorporated into the authenticity prediction device for telephone data.

[0232] The computer-readable medium carries one or more programs. When the one or more programs are executed by the authenticity prediction device for telephone data, the authenticity prediction device for telephone data executes the method shown in the above embodiment.

[0233] Computer program code for performing the operations of the present disclosure 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 cases involving a remote computer, the remote computer may be connected to the user's computer via 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).

[0234] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the module, program segment, or a part of 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 box can also occur in a different order than that marked in the accompanying drawings. For example, two 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 box in the block diagram and / or flowchart, and the combination of the boxes in the block diagram and / or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0235] The units involved in the embodiments described in this disclosure may be implemented in software or hardware. In some cases, the name of a unit does not limit the unit itself. For example, the first acquisition unit may also be described as a "unit for acquiring at least two Internet Protocol addresses."

[0236] The functions described above herein may be performed, at least in part, by one or more hardware logic components. For example, and without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip (SOCs), complex programmable logic devices (CPLDs), and the like.

[0237] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, 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 foregoing.

[0238] The authenticity prediction device for telephone data in the embodiment of the present application can be used to execute the technical solutions in the above-mentioned method embodiments of the present application. Its implementation principles and technical effects are similar and will not be repeated here.

[0239] An embodiment of the present application also provides a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are executed by a processor, they are used to implement any of the above-mentioned methods for predicting the authenticity of telephone data.

[0240] An embodiment of the present application also provides a computer program product, including a computer program, which, when executed by a processor, is used to implement any of the above-mentioned methods for predicting the authenticity of telephone data.

[0241] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0242] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0243] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, and the true scope and spirit of the present application are indicated by the following claims.

[0244] It should be understood that the present application is not limited to the exact structure described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A method for predicting the authenticity of telephone data, characterized in that: include: Acquire a telephone data set; wherein the telephone data set includes a plurality of telephone data and a business source corresponding to the telephone data; Generate a training data set and a test data set based on the telephone data set; wherein the training data set includes a plurality of training data, a business source corresponding to the training data, and a true label corresponding to the training data; Obtain the original authenticity prediction model; Taking the training data and the business source as input and the true label as output, the original authenticity prediction model is trained to obtain a trained authenticity prediction model; According to the test data set, the trained authenticity prediction model is iteratively trained using the maximum expectation algorithm until the preset iteration requirements are met, thereby obtaining a target prediction model; wherein, the target prediction model is used to output a target prediction result based on the input telephone data to be predicted and the business source corresponding to the telephone data to be predicted.

2. The method according to claim 1, characterized in that The method of iteratively training the trained authenticity prediction model using the maximum expectation algorithm according to the test data set until a preset iteration requirement is met to obtain a target prediction model includes: In the E-step calculation process of the maximum expectation algorithm, the trained authenticity prediction model is used to perform forward propagation processing on the test data set to obtain a test authenticity prediction probability; Determining a test set score based on the predicted probability of test authenticity; In the M-step calculation process of the maximum expectation algorithm, adjusting the model parameters and the judgment threshold of the trained authenticity prediction model according to the test set score; Continue to train the adjusted authenticity prediction model until the preset iteration requirements are met to obtain the target prediction model.

3. The method according to claim 2, characterized in that The adjusting the model parameters and the determination threshold of the trained authenticity prediction model according to the test set score includes: According to the test set score, adjusting the model parameters of the trained authenticity prediction model using a cross entropy loss function, an adaptive moment estimation optimizer, and a cosine annealing strategy; The model parameters of the trained authenticity prediction model are fixed, and the determination threshold of the trained authenticity prediction model is adjusted by a preset determination threshold loss function.

4. The method according to claim 2, characterized in that Before adjusting the model parameters and the decision threshold of the trained authenticity prediction model according to the test set score, the method further includes: Obtain a verification data set and determine the verification set score; Accordingly, adjusting the model parameters and decision thresholds of the trained authenticity prediction model according to the test set score includes: According to the test set score and the verification set score, the model parameters and the judgment threshold of the trained authenticity prediction model are adjusted.

5. The method according to claim 2, characterized in that The preset iteration requirements include: If the difference between the decision thresholds obtained from two consecutive trainings is less than the preset difference threshold, the iteration is stopped; and\or, If the total number of training times is greater than the preset training threshold, the iteration is stopped.

6. The method according to any one of claims 1 to 5, characterized in that Generating a training dataset and a test dataset based on the telephone dataset includes: Performing standardization and cleaning on the telephone data to obtain first preprocessed data; Performing structured processing on the first preprocessed data to obtain a preprocessed data set; Splitting the preprocessed data set according to a preset ratio to obtain a test data set and a training data sample; wherein the training data sample includes a plurality of training data and business sources corresponding to the training data; Obtaining true labels corresponding to the training data; A training data set is generated according to the true label and the training data samples.

7. The method according to any one of claims 1 to 5, characterized in that After iteratively training the trained authenticity prediction model using the maximum expectation algorithm according to the test data set until a preset iteration requirement is met and a target prediction model is obtained, the method further includes: Acquire the telephone data to be predicted and the service source corresponding to the telephone data to be predicted; A target prediction result is determined according to the target prediction model, the telephone data to be predicted, and the business source corresponding to the telephone data to be predicted.

8. A device for predicting the authenticity of telephone data, characterized in that: include: A first acquisition module is configured to acquire a telephone data set, wherein the telephone data set includes a plurality of telephone data and a service source corresponding to the telephone data; A generation module, configured to generate a training data set and a test data set based on the telephone data set; wherein the training data set includes a plurality of training data, a service source corresponding to the training data, and a true label corresponding to the training data; The second acquisition module is used to obtain the original authenticity prediction model; a training module, configured to take the training data and the business source as input and the true label as output, and perform training processing on the original authenticity prediction model to obtain a trained authenticity prediction model; An iteration module is used to iteratively train the trained authenticity prediction model based on the test data set using a maximum expectation algorithm until a preset iteration requirement is met, thereby obtaining a target prediction model; wherein, the target prediction model is used to output a target prediction result based on the input telephone data to be predicted and the business source corresponding to the telephone data to be predicted.

9. A device for predicting the authenticity of telephone data, characterized in that: include: a processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 7 when executed by a processor.

Images