Illegal outbound call identification method, device, equipment, medium and product

Abstract

The invention discloses an illegal call-out identification method. The method comprises the following steps: carrying out feature extraction on obtained multi-dimensional data of a to-be-identified number; inputting the extracted features into a pre-trained first recognition model and a pre-trained second recognition model, and outputting a first recognition result and a second recognition result; wherein the first identification model is used for calculating a probability value of illegal call-out of the to-be-identified number, and the second identification model is used for calculating a confidence coefficient of illegal call-out of the to-be-identified number; constructing a telephone traffic relation graph according to the extracted features, and calculating an evaluation score of illegal call-out of the to-be-identified number based on the telephone traffic relation graph; according to the first recognition result, the second recognition result and the evaluation score, calculating to obtain a comprehensive recognition score; and obtaining a violation identification result of the to-be-identified number based on the comprehensive identification score and a preset violation judgment rule. According to the method, the comprehensiveness, the accuracy and the reliability of illegal call-out recognition can be remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and in particular to a method, apparatus, device, medium, and product for identifying unauthorized outbound calls. Background Technology

[0002] With the increasing prevalence of telemarketing and telecommunications services, unauthorized outbound marketing activities are on the rise, seriously impacting user experience and corporate image. Unauthorized outbound calls refer to marketing to customers without using the telecommunications company's designated service hotline. These activities typically include business marketing, rating guidance, and phone follow-ups. Such behavior easily leads to customer resentment, trust crises, and complaints. Therefore, it is necessary to use technical means to capture and identify these unauthorized outbound call numbers to reduce customer complaints and improve user experience.

[0003] Existing identification technologies typically rely on manual rule design or a single model, resulting in low identification efficiency and difficulty in fully covering the multi-dimensional characteristics of illegal outbound calls, which can easily lead to misjudgments or missed judgments. Summary of the Invention

[0004] This invention provides a method for identifying unauthorized outbound calls, which can significantly improve the comprehensiveness, accuracy and reliability of unauthorized outbound call identification.

[0005] In a first aspect, embodiments of the present invention provide a method for identifying unauthorized outbound calls, comprising: Feature extraction is performed on the multidimensional data of the acquired number to be identified; wherein, the multidimensional data includes user feature data, call record data, operator log data, behavioral trajectory data, and home device network data; The extracted features are input into a pre-trained first recognition model and a second recognition model, and the first recognition result and the second recognition result are output; wherein, the first recognition model is used to calculate the probability value of the number to be identified making illegal outbound calls, and the second recognition model is used to calculate the confidence level of the number to be identified making illegal outbound calls; A call flow relationship graph is constructed based on the extracted features, and an evaluation score for the illegal outbound calls of the number to be identified is calculated based on the call flow relationship graph. A comprehensive recognition score is calculated based on the first recognition result, the second recognition result, and the evaluation score. Based on the comprehensive recognition score and the preset violation judgment rules, the violation recognition result of the number to be identified is obtained.

[0006] Furthermore, the method also includes: Data preprocessing is performed based on the data type of the multidimensional data, including: For the numerical data in the multidimensional data, perform missing value imputation, excessively sparse data deletion, outlier data detection and deletion, data binning, and numerical verification analysis; For the category type data in the multidimensional data, perform category noise data deletion, category mapping, and sparse category filling; For text-type data in the multidimensional data, perform invalid character deletion, abnormal character filtering, long text truncation, and text role identification; The label-type data in the multidimensional data is translated into text-type data.

[0007] Furthermore, the feature extraction from the acquired multidimensional data of the number to be identified includes: For the numerical data, category data, and text data in the multidimensional data, respectively, feature extraction algorithms are used to extract features to obtain numerical features, category features, and text features.

[0008] Furthermore, the first recognition model is a deep neural network model, and the training process of the first recognition model includes: A first initial sample set is obtained by acquiring structured data of several user numbers, preprocessing and extracting features from the structured data; wherein, the structured data includes user basic information, signaling data, call data and complaint content data; The first initial sample set is labeled to obtain the first training sample set; A deep neural network model is constructed, and the deep neural network model is trained using the first training sample set. The training objective is to output the probability value of the training sample belonging to an illegal outbound call.

[0009] Furthermore, the step of labeling the first initial sample set to obtain the first training sample set includes: The samples in the first initial sample set that belong to the category of illegal outbound calls are marked as positive samples; Samples that do not belong to the category of illegal outbound calls in the first initial sample set are labeled as negative samples.

[0010] Furthermore, training the deep neural network model using the first training sample set includes: The first training sample set is divided into a first training set and a first validation set according to a preset ratio; The deep neural network model is trained using the first training set, and the trained model is validated using the first validation set.

[0011] Furthermore, the second recognition model is a DeepSeek large model, and the training process of the second recognition model includes: Unstructured data of several user numbers is obtained, and the unstructured data is preprocessed; wherein, the unstructured data includes call text content, complaint information, and semantic information of social circles; The preprocessed unstructured data is filled into a preset Prompt template to obtain the second training sample set; The DeepSeek large model is trained using the second training sample set, with the training objective being to output the confidence level of training samples belonging to unauthorized outbound calls.

[0012] Furthermore, the step of constructing a call flow relationship graph based on the extracted features, and calculating the evaluation score for the illegal outbound calls of the number to be identified based on the call flow relationship graph, includes: Construct a call flow diagram, wherein the nodes of the call flow diagram are the number to be identified and all other user numbers that have call records with the number to be identified, and the edges of the call flow diagram are the call relationships between the number to be identified and other user numbers; Calculate the comprehensive risk score for each edge in the call relationship graph, and calculate the evaluation score for the illegal outbound calls of the number to be identified based on the comprehensive risk score; wherein, the comprehensive risk score is used to quantify the degree of illegal risk of the call relationship.

[0013] Furthermore, the node has node attributes, which include static attributes and dynamic attributes. The static attributes include user type, network access duration, occupation information, and home location. The user type includes individual type and enterprise type. The dynamic attributes include recent call frequency, call range, number of complaints, and call rejection rate.

[0014] Furthermore, the calculation of the comprehensive risk score for each edge in the call traffic relationship graph includes: Calculate the frequency risk, duration risk, spatiotemporal risk, and behavioral risk of each edge in the traffic relationship graph; Based on the frequency risk, duration risk, spatiotemporal risk, and behavioral risk, as well as the pre-set frequency weight, duration weight, spatiotemporal weight, and behavioral weight, calculate the comprehensive risk score for each edge.

[0015] Furthermore, the method also includes: Calculate the out-degree and in-degree of a node; wherein, the out-degree represents the average number of times the node's corresponding number calls different numbers per day, and the in-degree represents the average number of times the node's corresponding number is called by different numbers per day; When a node has an out-degree greater than a preset first out-degree threshold and an in-degree less than a preset first in-degree threshold, the node is identified as a high-risk calling node. When a node has an in-degree greater than the preset second in-degree threshold and an average call duration less than the preset call duration threshold, the node is identified as an abnormal called node.

[0016] Furthermore, the method also includes: When the overall risk score is greater than the preset risk score threshold, the corresponding edge will be identified as a high-risk call edge.

[0017] Furthermore, the method also includes: The edge removal operation in the traffic relationship graph includes: Filter out all nodes whose user numbers are official customer service numbers, and remove all edges corresponding to those nodes; Based on location trajectory overlap and home WiFi sharing records, call relationships belonging to home calls are filtered out, and the edges corresponding to the call relationships belonging to home calls are removed. Based on user occupation information and official business order records, filter out call relationships that belong to work calls, and remove the edges corresponding to the call relationships that belong to work calls. Remove edges corresponding to call relationships whose call duration is less than a preset call duration threshold.

[0018] Furthermore, the step of calculating the comprehensive recognition score based on the first recognition result, the second recognition result, and the evaluation score includes: Based on a pre-set first weight, second weight, and third weight, the first identification result, the second identification result, and the evaluation score are weighted and fused. The weighted fusion result is normalized to obtain a comprehensive recognition score.

[0019] Furthermore, the step of obtaining the violation identification result of the number to be identified based on the comprehensive identification score and the preset violation judgment rule includes: Pre-set thresholds for unauthorized outbound calls; When the comprehensive identification score is greater than the illegal outbound call threshold, the number to be identified is determined to be an illegal outbound call number; When the comprehensive identification score is less than or equal to the illegal outbound call threshold, the number to be identified is determined to be a non-illegal outbound call number.

[0020] Secondly, embodiments of the present invention provide a device for identifying unauthorized outbound calls, comprising: The feature extraction module is used to extract features from the multidimensional data of the acquired number to be identified; wherein, the multidimensional data includes user feature data, call record data, operator log data, behavioral trajectory data, and home device network data; The model recognition module is used to input the extracted features into a pre-trained first recognition model and a second recognition model, and output a first recognition result and a second recognition result; wherein, the first recognition model is used to calculate the probability value of the number to be identified making illegal outbound calls, and the second recognition model is used to calculate the confidence level of the number to be identified making illegal outbound calls; The relationship graph recognition module is used to construct a call flow relationship graph based on the extracted features, and calculate the evaluation score of the illegal outbound calls of the number to be identified based on the call flow relationship graph; The score calculation module is used to calculate a comprehensive recognition score based on the first recognition result, the second recognition result, and the evaluation score; The identification result determination module is used to obtain the violation identification result of the number to be identified based on the comprehensive identification score and the preset violation determination rules.

[0021] Furthermore, the device also includes a data preprocessing module for: Data preprocessing is performed based on the data type of the multidimensional data, including: For the numerical data in the multidimensional data, perform missing value imputation, excessively sparse data deletion, outlier data detection and deletion, data binning, and numerical verification analysis; For the category type data in the multidimensional data, perform category noise data deletion, category mapping, and sparse category filling; For text-type data in the multidimensional data, perform invalid character deletion, abnormal character filtering, long text truncation, and text role identification; The label-type data in the multidimensional data is translated into text-type data.

[0022] Furthermore, the feature extraction module is specifically used for: For the numerical data, category data, and text data in the multidimensional data, respectively, feature extraction algorithms are used to extract features to obtain numerical features, category features, and text features.

[0023] Furthermore, the first recognition model is a deep neural network model, and the training process of the first recognition model includes: A first initial sample set is obtained by acquiring structured data of several user numbers, preprocessing and extracting features from the structured data; wherein, the structured data includes user basic information, signaling data, call data and complaint content data; The first initial sample set is labeled to obtain the first training sample set; A deep neural network model is constructed, and the deep neural network model is trained using the first training sample set. The training objective is to output the probability value of the training sample belonging to an illegal outbound call.

[0024] Furthermore, the second recognition model is a DeepSeek large model, and the training process of the second recognition model includes: Unstructured data of several user numbers is obtained, and the unstructured data is preprocessed; wherein, the unstructured data includes call text content, complaint information, and semantic information of social circles; The preprocessed unstructured data is filled into a preset Prompt template to obtain the second training sample set; The DeepSeek large model is trained using the second training sample set, with the training objective being to output the confidence level of training samples belonging to unauthorized outbound calls.

[0025] Furthermore, the relationship graph recognition module is specifically used for: Construct a call flow diagram, wherein the nodes of the call flow diagram are the number to be identified and all other user numbers that have call records with the number to be identified, and the edges of the call flow diagram are the call relationships between the number to be identified and other user numbers; Calculate the comprehensive risk score for each edge in the call relationship graph, and calculate the evaluation score for the illegal outbound calls of the number to be identified based on the comprehensive risk score; wherein, the comprehensive risk score is used to quantify the degree of illegal risk of the call relationship.

[0026] Furthermore, the fraction calculation module is specifically used for: Based on a pre-set first weight, second weight, and third weight, the first identification result, the second identification result, and the evaluation score are weighted and fused. The weighted fusion result is normalized to obtain a comprehensive recognition score.

[0027] Furthermore, the recognition result determination module is specifically used for: Pre-set thresholds for unauthorized outbound calls; When the comprehensive identification score is greater than the illegal outbound call threshold, the number to be identified is determined to be an illegal outbound call number; When the comprehensive identification score is less than or equal to the illegal outbound call threshold, the number to be identified is determined to be a non-illegal outbound call number.

[0028] Thirdly, embodiments of the present invention provide an electronic device, comprising: Memory, used to store computer programs; A processor for executing the computer program; Wherein, when the processor executes the computer program, it implements the illegal outbound call identification method described in any of the first aspects above.

[0029] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program, which, when executed, implements the illegal outbound call identification method described in any of the first aspects above.

[0030] Fifthly, embodiments of the present invention provide a computer program product, including computer instructions, which, when executed by a processor, implement the illegal outbound call identification method described in any of the first aspects above.

[0031] Compared with existing technologies, the present invention provides a method for identifying unauthorized outbound calls, which has the following advantages: Feature extraction is performed on the multidimensional data of the number to be identified; wherein the multidimensional data includes user feature data, call record data, operator log data, behavioral trajectory data, and home device network data; the extracted features are input into a pre-trained first identification model and a second identification model, and a first identification result and a second identification result are output; wherein the first identification model is used to calculate the probability value of the unauthorized outbound call of the number to be identified, and the second identification model is used to calculate the confidence level of the unauthorized outbound call of the number to be identified; a call flow graph is constructed based on the extracted features, and an evaluation score for the unauthorized outbound call of the number to be identified is calculated based on the call flow graph; a comprehensive identification score is calculated based on the first identification result, the second identification result, and the evaluation score; and the unauthorized identification result of the number to be identified is obtained based on the comprehensive identification score and a preset violation judgment rule. The present invention can significantly improve the comprehensiveness, accuracy, and reliability of unauthorized outbound call identification. Attached Figure Description

[0032] To more clearly illustrate the technical features of the embodiments of the present invention, the drawings used in the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is a flowchart illustrating a method for identifying unauthorized outbound calls provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of an illegal outbound call identification device provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0036] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein is for the purpose of describing embodiments of the invention only and is not intended to limit the invention.

[0037] In a first aspect, embodiments of the present invention provide a method for identifying unauthorized outbound calls, see [link to previous document]. Figure 1 This is a flowchart illustrating one embodiment of a method for identifying unauthorized outbound calls provided by the present invention.

[0038] like Figure 1 As shown, the method includes the following steps: S1: Perform feature extraction on the multidimensional data of the acquired number to be identified; wherein, the multidimensional data includes user feature data, call record data, operator log data, behavioral trajectory data, and home device network data; The process involves acquiring multidimensional data on the number to be identified, including user characteristic data, call record data, operator log data, behavioral trajectory data, and home device network data. User characteristic data includes user type, network access duration, occupation information, and home location. Call record data includes caller and called numbers, call duration, frequency, time period, and rejection rate. Operator log data includes outbound call task type, base station information, and business registration records. Behavioral trajectory data includes base station switching frequency, call geographic distribution, and location overlap. Home device network data includes home WiFi sharing records and device association information. Features related to violation identification are extracted from the above multidimensional data to provide a data foundation for subsequent model input.

[0039] S2: Input the extracted features into the pre-trained first and second recognition models, and output the first and second recognition results; wherein, the first recognition model is used to calculate the probability value of the number to be identified making illegal outbound calls, and the second recognition model is used to calculate the confidence level of the number to be identified making illegal outbound calls; The extracted features are input into two pre-trained models. The first recognition model judges the probability of illegal outbound calls from a probability perspective and outputs the probability value of illegal outbound calls for the number to be identified. The higher the probability, the greater the suspicion of illegal behavior. The second recognition model measures the degree of violation with confidence and outputs the confidence score of illegal outbound calls for the number to be identified. The higher the confidence score, the greater the suspicion of semantic violation.

[0040] S3: Construct a call flow relationship graph based on the extracted features, and calculate the evaluation score of the illegal outbound calls of the number to be identified based on the call flow relationship graph; Based on the topological features extracted in step S1, a call relationship graph is constructed. By leveraging the call network's ability to mine complex node topological relationships, an evaluation score is output for any outbound calls made in violation of regulations by the numbers to be identified.

[0041] S4: Calculate the comprehensive recognition score based on the first recognition result, the second recognition result, and the evaluation score; The evaluation score of the dual-model results in step S2 and the relationship diagram in step S3 is fused into a unified comprehensive recognition score through preset weights, thereby achieving the complementarity of three-dimensional features.

[0042] S5: Based on the comprehensive identification score and the preset violation judgment rules, the violation identification result of the number to be identified is obtained.

[0043] By using preset violation judgment rules, the comprehensive identification score is converted into a binary classification result of violation / non-violation, so as to supervise illegal outbound call behavior.

[0044] In summary, this invention integrates the violation probability value of the first identification model, the violation confidence of the second identification model, and the violation assessment score of the call relationship graph, achieving comprehensive coverage of illegal outbound call behavior from three dimensions: behavioral characteristics, semantic intent, and correlation. Furthermore, by weighted fusion to obtain a comprehensive identification score, it integrates the advantages of each model, effectively making up for the limitations of identification by a single model or a single data dimension, and significantly improving the comprehensiveness, accuracy, and reliability of illegal outbound call identification.

[0045] In an optional implementation, the method further includes: Data preprocessing is performed based on the data type of the multidimensional data, including: For the numerical data in the multidimensional data, perform missing value imputation, excessively sparse data deletion, outlier data detection and deletion, data binning, and numerical verification analysis; For the category type data in the multidimensional data, perform category noise data deletion, category mapping, and sparse category filling; For text-type data in the multidimensional data, perform invalid character deletion, abnormal character filtering, long text truncation, and text role identification; The label-type data in the multidimensional data is translated into text-type data.

[0046] Specifically, to ensure the quality and consistency of multidimensional data, it is necessary to preprocess the multidimensional data. In this embodiment, different preprocessing methods are adopted for different types of multidimensional data. For numerical data, such as call duration, call frequency, base station location coordinates, network access duration, etc., missing values ​​of key indicators are filled in, low-value sparse data and abnormal data are detected and deleted directly, continuous values ​​are discretized by data binning, and numerical verification analysis is performed to verify the rationality and consistency of the data.

[0047] For categorized data, such as user type (individual / enterprise), complaint type (marketing harassment / false advertising), and geographic location, invalid noise data is removed, text categories are converted into machine-recognizable numerical codes, and low-frequency sparse categories are filled in.

[0048] For text-based data, such as transcripts of call content, descriptions of user complaints, and customer service scripts, filter out meaningless and abnormal characters, set a text length threshold, truncate excessively long texts, and mark the speaking roles in the dialogue text to clarify semantic attribution.

[0049] For multidimensional data with label types, such as base station tags (whether it is near a business hall), signaling tags (whether it is a high-frequency outbound call), and social circle tags (whether it has a wide social circle), there is a lack of semantic information. These need to be translated into natural language text to provide context for the subsequent semantic analysis of the DeepSeek large model.

[0050] This embodiment preprocesses the acquired multidimensional data to output three types of standardized data, providing high-quality data input for the subsequent first identification model, second identification model, and call flow relationship diagram.

[0051] In one optional implementation, the feature extraction of the acquired multidimensional data of the number to be identified includes: For the numerical data, category data, and text data in the multidimensional data, respectively, feature extraction algorithms are used to extract features to obtain numerical features, category features, and text features.

[0052] Specifically, based on the structural characteristics of different data types and combined with the core requirements of identifying illegal outbound calls, differentiated feature extraction is performed. For numerical data, methods such as uniform quantization, feature standardization, and statistical feature calculation can be used to eliminate the influence of units and enhance the comparability of numerical features. For categorical data, methods such as one-hot encoding, multi-hot encoding, and category mapping can be used to transform discrete text categories into machine-computable numerical vectors while preserving the logical relationships between categories. For textual data, methods such as text semantic embedding, n-gram technology, and keyword weight calculation can be used to quantify unstructured text and extract semantic-level clues of violations.

[0053] For example, core features extracted from call log data include call frequency, call duration, call time period, and call area; core features extracted from user feedback data include the number of complaints, complaint type, and user sentiment analysis; core features extracted from outbound call log data include outbound call task type, marketing content keywords, and outbound call number source; core features extracted from social circle information data include the number of social contacts, social circle dispersion, and reverse call rate; and core features extracted from call content data include the caller, call intent, and matching degree of illegal keywords.

[0054] This embodiment employs targeted feature extraction algorithms on different types of data in the multidimensional data of the numbers to be identified, which can comprehensively and accurately mine feature information related to illegal outbound calls in different types of data, providing strong support for the subsequent accurate identification of illegal outbound call numbers.

[0055] In one optional implementation, the first recognition model is a deep neural network model, and the training process of the first recognition model includes: A first initial sample set is obtained by acquiring structured data of several user numbers, preprocessing and extracting features from the structured data; wherein, the structured data includes user basic information, signaling data, call data and complaint content data; The first initial sample set is labeled to obtain the first training sample set; A deep neural network model is constructed, and the deep neural network model is trained using the first training sample set. The training objective is to output the probability value of the training sample belonging to an illegal outbound call.

[0056] Specifically, based on the requirements for identifying unauthorized outbound calls and the characteristics of data distribution, a deep neural network model (DNN) is selected as the primary identification model. The DNN model can extract high-order nonlinear features from multi-dimensional structured data and output the probability value of unauthorized outbound calls in the 0-1 range.

[0057] When training the DNN model, the structured data of several user numbers is first obtained as training data. The structured data includes basic user information (age, region, consumption level, network access duration, user type, occupation information, etc.), signaling data (base station location, base station handover frequency, number of people in the social circle, signaling interaction frequency, high-frequency outbound call markers, etc.), call data (caller / called number, call duration, call time period, call frequency, call area, called party rejection rate, etc.), and complaint content data (number of complaints, complaint type, complaint time, user emotion tags, etc.). The obtained structured data is preprocessed and features are extracted to obtain the first initial sample set.

[0058] Based on business rules, it is determined whether a sample is an illegal outbound call and it is labeled with a binary classification label to obtain the first training sample set. The first training sample set is used to train the DNN model. The trained DNN model outputs the probability value of illegal outbound calls for the input number to be identified. The higher the probability value, the greater the probability of illegality.

[0059] This embodiment uses a deep neural network model as the first identification model. By leveraging the powerful learning and fitting capabilities of deep neural networks, it can fully utilize users' multi-dimensional structured data to deeply mine potential patterns and features related to illegal outbound calls, accurately predict whether a number is making illegal outbound calls, and effectively improve the accuracy and reliability of illegal outbound call identification.

[0060] In one optional implementation, the step of labeling the first initial sample set to obtain the first training sample set includes: The samples in the first initial sample set that belong to the category of illegal outbound calls are marked as positive samples; Samples that do not belong to the category of illegal outbound calls in the first initial sample set are labeled as negative samples.

[0061] Specifically, based on the preset judgment rules, samples that belong to the violation of outbound calls are labeled as positive samples and samples that do not belong to the violation of outbound calls are labeled as negative samples through binary classification hard labeling, thus forming the first training sample set that can be directly used for DNN model training.

[0062] This embodiment labels the first initial sample set with positive and negative samples, enabling the model to effectively learn the differences in structured data features between illegal outbound calls and normal outbound calls based on these labeled data. In turn, it continuously optimizes its own parameters during the training process, thereby improving its ability to identify illegal outbound calls.

[0063] In one optional implementation, training the deep neural network model using the first training sample set includes: The first training sample set is divided into a first training set and a first validation set according to a preset ratio; The deep neural network model is trained using the first training set, and the trained model is validated using the first validation set.

[0064] Specifically, the first set of labeled training samples is split into a first training set and a first validation set. The first training set is used to update the weight parameters of the DNN model, allowing the model to learn the mapping relationship between feature vectors and violation labels. The first validation set is used to evaluate the model's performance on data not used in training, monitor the risk of overfitting during training, and guide the model's iterative optimization. The first training set is input into the DNN model in batches for iterative training. After each round of training, the loss value, precision, and recall of the training set are recorded. The first validation set is used to evaluate the model's performance, output the predicted violation probability values ​​for all samples, and save the model weights with the best performance on the validation set as the final training model.

[0065] The training process in this embodiment can improve the generalization ability and overall performance of the deep neural network model, enabling it to more accurately and reliably identify illegal outbound call numbers in practical applications.

[0066] In one optional implementation, the second recognition model is a DeepSeek large model, and the training process of the second recognition model includes: Unstructured data of several user numbers is obtained, and the unstructured data is preprocessed; wherein, the unstructured data includes call text content, complaint information, and semantic information of social circles; The preprocessed unstructured data is filled into a preset Prompt template to obtain the second training sample set; The DeepSeek large model is trained using the second training sample set, with the training objective being to output the confidence level of training samples belonging to unauthorized outbound calls.

[0067] Specifically, the DeepSeek large model, as a semantic-dimensional violation identification model, focuses on mining violation language and intent from unstructured data such as call text and complaint descriptions. By combining call content and semantic information, it prompts the large model to analyze what the subject of the dialogue is and what the intent is. The model can accurately identify the semantic features of violation outbound calls and output a violation confidence score in the range of 0-1. The higher the confidence score, the greater the probability of violation of the call behavior.

[0068] Unstructured data reflecting call semantics and user feedback is acquired, including call text content, complaint information, and social circle semantic information. The acquired data is preprocessed to ensure that it can be effectively integrated by the Prompt template. The call text and background semantic information of each sample are substituted into the Prompt template to generate complete model input text. Combined with manually labeled confidence labels, a second training sample set is formed. The DeepSeek large model is trained using the second training sample set to obtain the trained DeepSeek large model.

[0069] This embodiment selects the DeepSeek large model as the second recognition model. By leveraging the powerful unstructured data processing and understanding capabilities of the large model, it can accurately capture the semantic clues and features of illegal outbound calls in unstructured data, effectively improving the comprehensiveness and accuracy of illegal outbound call identification.

[0070] In one optional implementation, the step of constructing a call flow diagram based on the extracted features and calculating the evaluation score for the illegal outbound calls of the number to be identified based on the call flow diagram includes: Construct a call flow diagram, wherein the nodes of the call flow diagram are the number to be identified and all other user numbers that have call records with the number to be identified, and the edges of the call flow diagram are the call relationships between the number to be identified and other user numbers; Calculate the comprehensive risk score for each edge in the call relationship graph, and calculate the evaluation score for the illegal outbound calls of the number to be identified based on the comprehensive risk score; wherein, the comprehensive risk score is used to quantify the degree of illegal risk of the call relationship.

[0071] Specifically, the system acquires multi-source data required to construct the call flow graph, including user call flow data (calling / called numbers, call time, call frequency, call duration, call area), user behavior trajectory data (base station location, base station handover frequency, location overlap), user home network device data (shared WiFi connection records, device association duration), service subscription data (official service number marking, occupation information, service type), called call rejection records, user complaint markings, etc. Based on the acquired data, a call flow graph is constructed with user numbers as nodes and call relationships as edges. The node identifier is the user's mobile phone number, which includes the number to be identified and all other user numbers that have call records with the number to be identified. If there is a valid call record between the number to be identified A (caller) and number B (called), a directed edge A→B is constructed in the graph. Each edge is accompanied by multi-dimensional weight parameters (frequency, duration, spatiotemporal, behavioral weight). The weight value quantifies the violation risk of the call relationship. The edge weight is updated using a 24-hour sliding window. New call records trigger incremental graph calculations, eliminating the need for a full network reconstruction and ensuring real-time performance.

[0072] Calculate the comprehensive risk score for each edge in the call flow diagram, and then calculate the final evaluation score for any unauthorized outbound calls made by the number to be identified based on the comprehensive risk score.

[0073] This embodiment constructs a call relationship graph with the number to be identified and its associated numbers as nodes and call relationships as edges. It can comprehensively evaluate from the dimensions of number association and call risk, effectively improving the depth and breadth of identifying illegal outbound calls and enhancing the accuracy and reliability of the identification results.

[0074] In one optional implementation, the node has node attributes, which include static attributes and dynamic attributes. The static attributes include user type, network access duration, occupation information, and home location. The user type includes individual type and enterprise type. The dynamic attributes include recent call frequency, call range, number of complaints, and call rejection rate.

[0075] Specifically, node attributes are key dimensions for distinguishing between compliant and non-compliant outbound call characteristics of users. These include static and dynamic attributes. Static attributes are the long-term stable characteristics of a user's number that do not change with short-term call behavior, including user type (individual or enterprise), network access duration, occupation information, and home location. Dynamic attributes are the statistical characteristics of a user's recent call behavior, calculated based on call data from a 24-hour sliding window, reflecting changes in user call behavior in real time, including recent call frequency, call range, number of complaints, and call rejection rate.

[0076] In one optional implementation, calculating the comprehensive risk score for each edge in the traffic relationship graph includes: Calculate the frequency risk, duration risk, spatiotemporal risk, and behavioral risk of each edge in the traffic relationship graph; Based on the frequency risk, duration risk, spatiotemporal risk, and behavioral risk, as well as the pre-set frequency weight, duration weight, spatiotemporal weight, and behavioral weight, calculate the comprehensive risk score for each edge.

[0077] Specifically, for each directed call edge in the call flow diagram, four single-dimensional indicators are calculated: frequency risk, duration risk, spatiotemporal risk, and behavioral risk. Frequency risk reflects the frequency of calls from the calling number to the called number; the higher the frequency, the greater the suspicion of mass marketing. Duration risk reflects the degree of abnormality in call duration; short-duration calls usually correspond to violations such as robot outbound calls and harassing calls. Spatiotemporal risk integrates two sub-indicators: "proportion of calls from the same base station" and "proportion of calls across regions," reflecting the abnormality of the geographical distribution of outbound call behavior. Behavioral risk integrates two sub-indicators: "called rejection rate" and "complaint marking," reflecting the degree of aversion of the called user to the call.

[0078] After obtaining four single-dimensional risk indicators, the comprehensive risk score of each edge is calculated by weighted summation based on the pre-set dimension weight coefficients. The comprehensive risk score ranges from 0 to 1. The higher the score, the higher the risk of violation for that call edge. The allocation of weight coefficients is based on the contribution of each dimension to the determination of illegal outbound calls, and is determined by combining business scenarios and historical data verification. No specific limitations are made here.

[0079] This embodiment calculates the risk of each edge in the call relationship graph in four dimensions: frequency, duration, spatiotemporal, and behavior. Based on the pre-set risk weights of various risks, it comprehensively calculates the overall risk score of each edge. This allows for a comprehensive and detailed quantification of the degree of violation risk in call relationships, accurately captures the impact of different risk factors on unauthorized outbound calls, and effectively improves the accuracy and scientific nature of the assessment.

[0080] In an optional implementation, the method further includes: Calculate the out-degree and in-degree of a node; wherein, the out-degree represents the average number of times the node's corresponding number calls different numbers per day, and the in-degree represents the average number of times the node's corresponding number is called by different numbers per day; When a node has an out-degree greater than a preset first out-degree threshold and an in-degree less than a preset first in-degree threshold, the node is identified as a high-risk calling node. When a node has an in-degree greater than the preset second in-degree threshold and an average call duration less than the preset call duration threshold, the node is identified as an abnormal called node.

[0081] Specifically, out-degree and in-degree are core indicators for characterizing node call behavior patterns. Out-degree represents the average number of different called numbers actively called by the node's corresponding number within the sliding window per day, reflecting the number's active outbound call coverage. The higher the out-degree, the greater the suspicion of mass outbound calls. In-degree represents the average number of different calling numbers the node's corresponding number is called by within the sliding window per day, reflecting the number's passive call coverage. The higher the in-degree, the more users actively contact the number.

[0082] High-risk calling nodes are the core source of illegal outbound calls. Typical characteristics include a wide range of active outbound calls and few passive calls, corresponding to illegal behaviors such as bulk marketing and robot harassment. When the outbound degree of a node is greater than the preset first outbound degree threshold and the inbound degree is less than the preset first inbound degree threshold, the node is identified as a high-risk calling node. For example, the first inbound degree threshold can be 10% of the node's outbound degree.

[0083] Abnormal called nodes are usually associated with scenarios such as receiving harassing calls or being called in bulk. When a node's in-degree is greater than a preset second in-degree threshold and its average call duration is less than a preset call duration threshold, the node is identified as an abnormal called node.

[0084] This embodiment can accurately identify high-risk calling nodes and abnormal called nodes by calculating the out-degree and in-degree of nodes. This helps to quickly locate key numbers that may have illegal outbound calling behavior from the call flow diagram, discover potential risks in advance, and provide a clear direction for subsequent targeted prevention and control of illegal outbound calling.

[0085] In an optional implementation, the method further includes: When the overall risk score is greater than the preset risk score threshold, the corresponding edge will be identified as a high-risk call edge.

[0086] Specifically, risk score thresholds are pre-set to define the boundaries between compliant and high-risk call edges. When the overall risk score of an edge exceeds the risk score threshold, that edge is identified as a high-risk call edge. Combining the identification results of high-risk call edges with node characteristics can provide a core basis for the node violation assessment and subsequent handling of the call flow diagram.

[0087] This embodiment can quickly filter out call connections with a high risk of violation from numerous call relationship edges, and accurately locate call links with potential illegal outbound calls.

[0088] In an optional implementation, the method further includes: The edge removal operation in the traffic relationship graph includes: Filter out all nodes whose user numbers are official customer service numbers, and remove all edges corresponding to those nodes; Based on location trajectory overlap and home WiFi sharing records, call relationships belonging to home calls are filtered out, and the edges corresponding to the call relationships belonging to home calls are removed. Based on user occupation information and official business order records, filter out call relationships that belong to work calls, and remove the edges corresponding to the call relationships that belong to work calls. Remove edges corresponding to call relationships whose call duration is less than a preset call duration threshold.

[0089] Specifically, before calculating the comprehensive risk score, it is necessary to remove compliant call edges that are irrelevant to the determination of illegal outbound calls. For call edges associated with officially certified customer service numbers, all related edges are directly removed to avoid including compliant customer service calls in the risk assessment. For normal calls within the user's home, they are filtered and removed based on location and device association data to reduce the interference of calls from relatives and friends on the determination of violations. For compliant work calls based on the user's professional identity, they are filtered and removed in combination with professional information and business records. For short-duration misdialing edges without actual call content, they are filtered and removed based on call duration thresholds.

[0090] This embodiment improves the accuracy and relevance of subsequent violation assessments by eliminating non-violation-related edges such as official customer service calls, family calls, work calls, and short-duration misdialed calls, focusing on call relationships that may involve illegal outbound calls. This makes the results of illegal outbound call identification more reliable.

[0091] In one optional implementation, calculating the comprehensive recognition score based on the first recognition result, the second recognition result, and the evaluation score includes: Based on a pre-set first weight, second weight, and third weight, the first identification result, the second identification result, and the evaluation score are weighted and fused. The weighted fusion result is normalized to obtain a comprehensive recognition score.

[0092] Specifically, the probability value of the first identification model (DNN), the confidence level of the second identification model (DeepSeek large model), and the evaluation score of the call relationship graph are fused together to output a unified comprehensive identification score. The first weight, the second weight, and the third weight are preset to ensure that the sum of the weights is 1. The allocation of weight coefficients should be based on the business contribution of each model in violation judgment, combined with historical data verification and manual evaluation, and no specific restrictions are made here.

[0093] By normalizing the results, the weighted fusion is adjusted to the 0-1 range. The higher the score, the greater the likelihood that the number is making illegal outbound calls.

[0094] This embodiment can comprehensively consider the impact of structured data, unstructured data, and call relationship information on the judgment of illegal outbound calls, making the comprehensive identification more comprehensive and reasonable, and providing a more reliable basis for accurately judging illegal outbound calls.

[0095] In one optional implementation, obtaining the violation identification result of the number to be identified based on the comprehensive identification score and the preset violation judgment rule includes: Pre-set thresholds for unauthorized outbound calls; When the comprehensive identification score is greater than the illegal outbound call threshold, the number to be identified is determined to be an illegal outbound call number; When the comprehensive identification score is less than or equal to the illegal outbound call threshold, the number to be identified is determined to be a non-illegal outbound call number.

[0096] Specifically, a threshold for illegal outbound calls is pre-set to define the boundary between illegal and non-illegal calls. When the comprehensive identification score is greater than the illegal outbound call threshold, the number to be identified is determined to be an illegal outbound call number. When the comprehensive identification score is less than or equal to the illegal outbound call threshold, the number to be identified is determined to be a non-illegal outbound call number.

[0097] This embodiment uses a pre-defined threshold for illegal outbound calls to clearly and quickly determine whether a number to be identified is an illegal outbound call number based on a comparison between the comprehensive identification score and the threshold.

[0098] Secondly, embodiments of the present invention provide a device for identifying unauthorized outbound calls, see [link to relevant documentation]. Figure 2 This is a schematic diagram of one embodiment of an illegal outbound call identification device provided by the present invention.

[0099] like Figure 2 As shown, the device includes: The feature extraction module 21 is used to extract features from the multidimensional data of the acquired number to be identified; wherein, the multidimensional data includes user feature data, call record data, operator log data, behavioral trajectory data, and home device network data; The model recognition module 22 is used to input the extracted features into a pre-trained first recognition model and a second recognition model, and output a first recognition result and a second recognition result; wherein, the first recognition model is used to calculate the probability value of the number to be identified making illegal outbound calls, and the second recognition model is used to calculate the confidence level of the number to be identified making illegal outbound calls; The relationship graph recognition module 23 is used to construct a call flow relationship graph based on the extracted features, and calculate the evaluation score of the illegal outbound calls of the number to be identified based on the call flow relationship graph; The score calculation module 24 is used to calculate a comprehensive recognition score based on the first recognition result, the second recognition result, and the evaluation score; The identification result determination module 25 is used to obtain the violation identification result of the number to be identified based on the comprehensive identification score and the preset violation determination rules.

[0100] In an optional embodiment, the apparatus further includes a data preprocessing module for: Data preprocessing is performed based on the data type of the multidimensional data, including: For the numerical data in the multidimensional data, perform missing value imputation, excessively sparse data deletion, outlier data detection and deletion, data binning, and numerical verification analysis; For the category type data in the multidimensional data, perform category noise data deletion, category mapping, and sparse category filling; For text-type data in the multidimensional data, perform invalid character deletion, abnormal character filtering, long text truncation, and text role identification; The label-type data in the multidimensional data is translated into text-type data.

[0101] In one optional implementation, the feature extraction module is specifically used for: For the numerical data, category data, and text data in the multidimensional data, respectively, feature extraction algorithms are used to extract features to obtain numerical features, category features, and text features.

[0102] In one optional implementation, the first recognition model is a deep neural network model, and the training process of the first recognition model includes: A first initial sample set is obtained by acquiring structured data of several user numbers, preprocessing and extracting features from the structured data; wherein, the structured data includes user basic information, signaling data, call data and complaint content data; The first initial sample set is labeled to obtain the first training sample set; A deep neural network model is constructed, and the deep neural network model is trained using the first training sample set. The training objective is to output the probability value of the training sample belonging to an illegal outbound call.

[0103] In one optional implementation, the second recognition model is a DeepSeek large model, and the training process of the second recognition model includes: Unstructured data of several user numbers is obtained, and the unstructured data is preprocessed; wherein, the unstructured data includes call text content, complaint information, and semantic information of social circles; The preprocessed unstructured data is filled into a preset Prompt template to obtain the second training sample set; The DeepSeek large model is trained using the second training sample set, with the training objective being to output the confidence level of training samples belonging to unauthorized outbound calls.

[0104] In one optional implementation, the relationship graph recognition module is specifically used for: Construct a call flow diagram, wherein the nodes of the call flow diagram are the number to be identified and all other user numbers that have call records with the number to be identified, and the edges of the call flow diagram are the call relationships between the number to be identified and other user numbers; Calculate the comprehensive risk score for each edge in the call relationship graph, and calculate the evaluation score for the illegal outbound calls of the number to be identified based on the comprehensive risk score; wherein, the comprehensive risk score is used to quantify the degree of illegal risk of the call relationship.

[0105] In one optional implementation, the fraction calculation module is specifically used for: Based on a pre-set first weight, second weight, and third weight, the first identification result, the second identification result, and the evaluation score are weighted and fused. The weighted fusion result is normalized to obtain a comprehensive recognition score.

[0106] In one optional implementation, the identification result determination module is specifically used for: Pre-set thresholds for unauthorized outbound calls; When the comprehensive identification score is greater than the illegal outbound call threshold, the number to be identified is determined to be an illegal outbound call number; When the comprehensive identification score is less than or equal to the illegal outbound call threshold, the number to be identified is determined to be a non-illegal outbound call number.

[0107] It should be noted that the illegal outbound call identification device provided in this embodiment of the invention is used to execute all the process steps of the illegal outbound call identification method in the above embodiment. The working principle and beneficial effect of the two are one-to-one, so they will not be described again.

[0108] Thirdly, embodiments of the present invention provide an electronic device, see [link to previous document]. Figure 3 The diagram shown is a structural schematic of an electronic device provided in an embodiment of the present invention.

[0109] like Figure 3 As shown, the device includes: Memory 31 is used to store computer programs; Processor 32 is used to execute the computer program; When the processor 32 executes the computer program, it implements the illegal outbound call identification method as described in any of the above embodiments.

[0110] For example, the computer program may be divided into one or more modules / units, which are stored in the memory 31 and executed by the processor 32 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0111] The processor 32 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0112] The memory 31 can be used to store the computer programs and / or modules. The processor 32 implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory 31 and calling the data stored in the memory 31. The memory 31 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 31 may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital card (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0113] It should be noted that the aforementioned electronic devices include, but are not limited to, processors and memory, as will be understood by those skilled in the art. Figure 3 The structural diagram is merely an example of the electronic device described above and does not constitute a limitation on the electronic device. It may include more components than shown in the diagram, or combine certain components, or use different components.

[0114] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed, implements the illegal outbound call identification method described in any of the above embodiments.

[0115] It should be understood that the implementation of all or part of the above-described method for identifying unauthorized outbound calls can also be accomplished by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the above-described method for identifying unauthorized outbound calls. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0116] Fifthly, embodiments of this application also provide a computer program product, which is stored in a storage medium and executed by at least one processor to implement the illegal outbound call identification method described in any of the above embodiments.

[0117] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. It should be noted that, for those skilled in the art, several equivalent obvious modifications and / or equivalent substitutions can be made without departing from the technical principles of the present invention, and these obvious modifications and / or equivalent substitutions should also be considered within the scope of protection of the present invention.

Claims

1. A method for identifying unauthorized outbound calls, characterized in that, include: Feature extraction is performed on the multidimensional data of the acquired number to be identified; wherein, the multidimensional data includes user feature data, call record data, operator log data, behavioral trajectory data, and home device network data; The extracted features are input into a pre-trained first recognition model and a second recognition model, and the first recognition result and the second recognition result are output; wherein, the first recognition model is used to calculate the probability value of the number to be identified making illegal outbound calls, and the second recognition model is used to calculate the confidence level of the number to be identified making illegal outbound calls; A call flow relationship graph is constructed based on the extracted features, and an evaluation score for the illegal outbound calls of the number to be identified is calculated based on the call flow relationship graph. A comprehensive recognition score is calculated based on the first recognition result, the second recognition result, and the evaluation score. Based on the comprehensive recognition score and the preset violation judgment rules, the violation recognition result of the number to be identified is obtained.

2. The method for identifying unauthorized outbound calls as described in claim 1, characterized in that, The method further includes: Data preprocessing is performed based on the data type of the multidimensional data, including: For the numerical data in the multidimensional data, perform missing value imputation, excessively sparse data deletion, outlier data detection and deletion, data binning, and numerical verification analysis; For the category type data in the multidimensional data, perform category noise data deletion, category mapping, and sparse category filling; For text-type data in the multidimensional data, perform invalid character deletion, abnormal character filtering, long text truncation, and text role identification; The label-type data in the multidimensional data is translated into text-type data.

3. The method for identifying unauthorized outbound calls as described in claim 1, characterized in that, The feature extraction of the acquired multidimensional data of the number to be identified includes: For the numerical data, category data, and text data in the multidimensional data, respectively, feature extraction algorithms are used to extract features to obtain numerical features, category features, and text features.

4. The method for identifying unauthorized outbound calls as described in claim 1, characterized in that, The first recognition model is a deep neural network model, and the training process of the first recognition model includes: A first initial sample set is obtained by acquiring structured data of several user numbers, preprocessing and extracting features from the structured data; wherein, the structured data includes user basic information, signaling data, call data and complaint content data; The first initial sample set is labeled to obtain the first training sample set; A deep neural network model is constructed, and the deep neural network model is trained using the first training sample set. The training objective is to output the probability value of the training sample belonging to an illegal outbound call.

5. The method for identifying unauthorized outbound calls as described in claim 4, characterized in that, The step of labeling the first initial sample set to obtain the first training sample set includes: The samples in the first initial sample set that belong to the category of illegal outbound calls are marked as positive samples; Samples that do not belong to the category of illegal outbound calls in the first initial sample set are labeled as negative samples.

6. The method for identifying unauthorized outbound calls as described in claim 4, characterized in that, The step of training the deep neural network model using the first training sample set includes: The first training sample set is divided into a first training set and a first validation set according to a preset ratio; The deep neural network model is trained using the first training set, and the trained model is validated using the first validation set.

7. The method for identifying unauthorized outbound calls as described in claim 1, characterized in that, The second recognition model is a DeepSeek large model, and the training process of the second recognition model includes: Unstructured data of several user numbers is obtained, and the unstructured data is preprocessed; wherein, the unstructured data includes call text content, complaint information, and semantic information of social circles; The preprocessed unstructured data is filled into a preset Prompt template to obtain the second training sample set; The DeepSeek large model is trained using the second training sample set, with the training objective being to output the confidence level of training samples belonging to unauthorized outbound calls.

8. The method for identifying unauthorized outbound calls as described in claim 1, characterized in that, The step of constructing a call flow relationship graph based on the extracted features and calculating the evaluation score for the illegal outbound calls of the number to be identified based on the call flow relationship graph includes: Construct a call flow diagram, wherein the nodes of the call flow diagram are the number to be identified and all other user numbers that have call records with the number to be identified, and the edges of the call flow diagram are the call relationships between the number to be identified and other user numbers; Calculate the comprehensive risk score for each edge in the call relationship graph, and calculate the evaluation score for the illegal outbound calls of the number to be identified based on the comprehensive risk score; wherein, the comprehensive risk score is used to quantify the degree of illegal risk of the call relationship.

9. The method for identifying unauthorized outbound calls as described in claim 8, characterized in that, The node has node attributes, which include static attributes and dynamic attributes. The static attributes include user type, network access duration, occupation information, and home location. The user type includes individual type and enterprise type. The dynamic attributes include recent call frequency, call range, number of complaints, and call rejection rate.

10. The method for identifying unauthorized outbound calls as described in claim 8, characterized in that, The calculation of the comprehensive risk score for each edge in the traffic relationship graph includes: Calculate the frequency risk, duration risk, spatiotemporal risk, and behavioral risk of each edge in the traffic relationship graph; Based on the frequency risk, duration risk, spatiotemporal risk, and behavioral risk, as well as the pre-set frequency weight, duration weight, spatiotemporal weight, and behavioral weight, calculate the comprehensive risk score for each edge.

11. The method for identifying unauthorized outbound calls as described in claim 8, characterized in that, The method further includes: Calculate the out-degree and in-degree of a node; wherein, the out-degree represents the average number of times the node's corresponding number calls different numbers per day, and the in-degree represents the average number of times the node's corresponding number is called by different numbers per day; When a node has an out-degree greater than a preset first out-degree threshold and an in-degree less than a preset first in-degree threshold, the node is identified as a high-risk calling node. When a node has an in-degree greater than the preset second in-degree threshold and an average call duration less than the preset call duration threshold, the node is identified as an abnormal called node.

12. The method for identifying unauthorized outbound calls as described in claim 8, characterized in that, The method further includes: When the overall risk score is greater than the preset risk score threshold, the corresponding edge will be identified as a high-risk call edge.

13. The method for identifying unauthorized outbound calls as described in claim 8, characterized in that, The method further includes: The edge removal operation in the traffic relationship graph includes: Filter out all nodes whose user numbers are official customer service numbers, and remove all edges corresponding to those nodes; Based on location trajectory overlap and home WiFi sharing records, call relationships belonging to home calls are filtered out, and the edges corresponding to the call relationships belonging to home calls are removed. Based on user occupation information and official business order records, filter out call relationships that belong to work calls, and remove the edges corresponding to the call relationships that belong to work calls. Remove edges corresponding to call relationships whose call duration is less than a preset call duration threshold.

14. The method for identifying unauthorized outbound calls as described in claim 1, characterized in that, The step of calculating a comprehensive recognition score based on the first recognition result, the second recognition result, and the evaluation score includes: Based on a pre-set first weight, second weight, and third weight, the first identification result, the second identification result, and the evaluation score are weighted and fused. The weighted fusion result is normalized to obtain a comprehensive recognition score.

15. The method for identifying unauthorized outbound calls as described in claim 1, characterized in that, The method of obtaining the violation identification result of the number to be identified based on the comprehensive identification score and the preset violation judgment rule includes: Pre-set thresholds for unauthorized outbound calls; When the comprehensive identification score is greater than the illegal outbound call threshold, the number to be identified is determined to be an illegal outbound call number; When the comprehensive identification score is less than or equal to the illegal outbound call threshold, the number to be identified is determined to be a non-illegal outbound call number.

16. A device for identifying unauthorized outbound calls, characterized in that, include: The feature extraction module is used to extract features from the multidimensional data of the acquired number to be identified; wherein, the multidimensional data includes user feature data, call record data, operator log data, behavioral trajectory data, and home device network data; The model recognition module is used to input the extracted features into a pre-trained first recognition model and a second recognition model, and output a first recognition result and a second recognition result; wherein, the first recognition model is used to calculate the probability value of the number to be identified making illegal outbound calls, and the second recognition model is used to calculate the confidence level of the number to be identified making illegal outbound calls; The relationship graph recognition module is used to construct a call flow relationship graph based on the extracted features, and calculate the evaluation score of the illegal outbound calls of the number to be identified based on the call flow relationship graph; The score calculation module is used to calculate a comprehensive recognition score based on the first recognition result, the second recognition result, and the evaluation score; The identification result determination module is used to obtain the violation identification result of the number to be identified based on the comprehensive identification score and the preset violation determination rules.

17. The illegal outbound call identification device as described in claim 16, characterized in that, The device further includes a data preprocessing module for: Data preprocessing is performed based on the data type of the multidimensional data, including: For the numerical data in the multidimensional data, perform missing value imputation, excessively sparse data deletion, outlier data detection and deletion, data binning, and numerical verification analysis; For the category type data in the multidimensional data, perform category noise data deletion, category mapping, and sparse category filling; For text-type data in the multidimensional data, perform invalid character deletion, abnormal character filtering, long text truncation, and text role identification; The label-type data in the multidimensional data is translated into text-type data.

18. The illegal outbound call identification device as described in claim 16, characterized in that, The feature extraction module is specifically used for: For the numerical data, category data, and text data in the multidimensional data, respectively, feature extraction algorithms are used to extract features to obtain numerical features, category features, and text features.

19. The illegal outbound call identification device as described in claim 16, characterized in that, The first recognition model is a deep neural network model, and the training process of the first recognition model includes: A first initial sample set is obtained by acquiring structured data of several user numbers, preprocessing and extracting features from the structured data; wherein, the structured data includes user basic information, signaling data, call data and complaint content data; The first initial sample set is labeled to obtain the first training sample set; A deep neural network model is constructed, and the deep neural network model is trained using the first training sample set. The training objective is to output the probability value of the training sample belonging to an illegal outbound call.

20. The illegal outbound call identification device as described in claim 16, characterized in that, The second recognition model is a DeepSeek large model, and the training process of the second recognition model includes: Unstructured data of several user numbers is obtained, and the unstructured data is preprocessed; wherein, the unstructured data includes call text content, complaint information, and semantic information of social circles; The preprocessed unstructured data is filled into a preset Prompt template to obtain the second training sample set; The DeepSeek large model is trained using the second training sample set, with the training objective being to output the confidence level of training samples belonging to unauthorized outbound calls.

21. The illegal outbound call identification device as described in claim 16, characterized in that, The relationship graph recognition module is specifically used for: Construct a call flow diagram, wherein the nodes of the call flow diagram are the number to be identified and all other user numbers that have call records with the number to be identified, and the edges of the call flow diagram are the call relationships between the number to be identified and other user numbers; Calculate the comprehensive risk score for each edge in the call relationship graph, and calculate the evaluation score for the illegal outbound calls of the number to be identified based on the comprehensive risk score; wherein, the comprehensive risk score is used to quantify the degree of illegal risk of the call relationship.

22. The illegal outbound call identification device as described in claim 16, characterized in that, The fraction calculation module is specifically used for: Based on a pre-set first weight, second weight, and third weight, the first identification result, the second identification result, and the evaluation score are weighted and fused. The weighted fusion result is normalized to obtain a comprehensive recognition score.

23. The illegal outbound call identification device as described in claim 16, characterized in that, The recognition result determination module is specifically used for: Pre-set thresholds for unauthorized outbound calls; When the comprehensive identification score is greater than the illegal outbound call threshold, the number to be identified is determined to be an illegal outbound call number; When the comprehensive identification score is less than or equal to the illegal outbound call threshold, the number to be identified is determined to be a non-illegal outbound call number.

24. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program; The processor executes the computer program to implement the illegal outbound call identification method as described in any one of claims 1 to 15.

25. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, implements the unauthorized outbound call identification method as described in any one of claims 1 to 15.

26. A computer program product, characterized in that, It includes computer instructions that, when executed by a processor, implement the unauthorized outbound call identification method as described in any one of claims 1 to 15.

Images