A machine learning-based telephone triage method, system, and storage medium
By constructing a sample feature library of both malicious and legitimate phone numbers, and using machine learning algorithms to build a decision tree and periodically update the feature combinations, the accuracy and timeliness issues of malicious call identification in existing technologies are solved, achieving efficient malicious call detection.
Patent Information
- Application Number
- CN202510049291.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-13
- Publication Date
- 2026-05-15
- Estimated Expiration
- 2045-01-13
AI Technical Summary
Existing methods for attracting traffic via telephone cannot accurately and efficiently identify harmful calls, nor can they be updated in a timely manner to adapt to changes in the characteristics of harmful calls.
A sample feature library of harmful and normal numbers is constructed. A decision tree is built using machine learning algorithms to select the optimal feature combination and segmentation threshold. The decision tree is used to identify the calling numbers, and the feature combination is updated periodically to adapt to changes.
It improves the accuracy of malicious call identification, reduces false positives and false negatives, and ensures that malicious calls can still be effectively detected when the characteristics of malicious numbers change.
Smart Images
Figure CN119892984B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of telecommunications network security technology, and in particular to a method, system, and storage medium for redirecting telephone traffic based on machine learning. Background Technology
[0002] Harmful call referral refers to filtering high-risk calls from the entire call log based on the caller ID or called number, and then sending these calls to a call intent analysis platform for analysis to detect fraudulent activity. If fraudulent activity is detected, the call is promptly controlled to prevent victims from being scammed. Current methods for analyzing harmful call referrals mainly include: blacklist / whitelist strategies, rule-based strategies, and strategies involving manually analyzing and issuing graylist numbers.
[0003] Among the various methods, blacklist / whitelist-based lead generation strategies offer advantages in terms of simplicity and speed, but suffer from drawbacks such as incomplete coverage and untimely updates. Rule-based lead generation strategies allow for flexible configuration of rules based on the latest harmful call patterns, but their broad coverage makes it difficult to filter out calls from legitimate users, leading to wasted lead generation resources. Manually analyzed and distributed graylist lead generation strategies can accurately identify harmful call characteristics and distribute lead numbers, but these manual graylist analysis mechanisms have long analysis cycles, require highly experienced personnel, and cannot be automatically updated in a timely manner. Effective graylist lead generation strategies developed initially may become less effective as harmful call characteristics change. Therefore, a precise and efficient solution is needed to construct intelligent call generation strategies, achieve dynamic call monitoring, and proactively detect harmful calls. Summary of the Invention
[0004] Purpose of the invention: The purpose of this invention is to provide a method, system, and storage medium for telephone lead generation based on machine learning, which can effectively identify harmful telephone calls.
[0005] Technical Solution: To achieve the above objectives, the present invention provides a machine learning-based telephone lead generation method, comprising the following steps:
[0006] S1: Construct a sample feature library including both harmful and normal phone numbers;
[0007] S2: Preprocess the feature data in the sample feature library and divide it into a feature training set and a feature test set;
[0008] S3: Construct a decision tree based on the feature training set;
[0009] S4: Based on the feature test set, filter the decision rules in the decision tree to obtain the feature combination that performs best on the test set and the segmentation threshold corresponding to the feature combination;
[0010] S5: Collect new batches of call numbers and obtain the feature data of the call numbers. Based on the optimal feature combination and the segmentation threshold corresponding to the feature combination, identify the feature data of the call numbers to predict whether the call number is a harmful number or a normal number.
[0011] S6: Repeat steps S1-S5 to periodically update the optimal feature combination and its corresponding segmentation threshold.
[0012] The method for constructing the sample feature library in S1 is as follows: the case system is connected through the system interface or file synchronization, and the calling and called numbers of harmful calls are extracted from the case system as harmful number samples and normal number samples, respectively, to dynamically construct the sample feature library. The sample feature library includes basic features based on communication behavior and statistical features obtained by analyzing the basic features.
[0013] The preprocessing methods in S2 include data cleaning and missing value handling.
[0014] The decision tree generated in S3 contains a series of paths from the root node to the leaf node. Each path represents a decision rule, and each decision rule contains multiple feature combinations. Each feature combination has a corresponding segmentation threshold.
[0015] The decision tree construction process includes:
[0016] S3.1: Feature selection: Select the most discriminative feature from the feature training set as the root node or internal node;
[0017] S3.2: Threshold selection: Determine a segmentation threshold for the most discriminative feature selected;
[0018] S3.3: Recursive splitting. Starting from the root node, based on the selected features and segmentation threshold, the feature training set is split into several internal nodes. For each internal node, the most discriminative feature and corresponding segmentation threshold are selected from the remaining feature training set for further splitting. This feature selection and threshold selection process is repeated until the stopping condition is met. For the child nodes after the split of each internal node, feature selection and threshold selection are further performed from the remaining feature training set, and this process is repeated until the stopping condition is met.
[0019] S3.4: Optimize decision tree pruning.
[0020] Among them, the feature with the lowest Gini index is selected as the most discriminative feature. The formula for calculating the Gini index is as follows:
[0021]
[0022] Where m is the number of samples at a node, n is the number of categories, and c is the number of samples in category i. i .
[0023] The stopping conditions are reaching a predetermined maximum tree depth, or the number of node samples is less than a certain set threshold, or the purity of samples within a node reaches a certain set standard.
[0024] The selection method in S4 is as follows: the selection method of cross-validation selects the feature combination that performs best on multiple test sets as the optimal feature combination. Each test data starts from the root node of the decision tree and selects a path according to the features of the test data until it reaches a certain leaf node. The leaf node gives the final classification prediction result, such as normal or harmful. Based on the actual label of the test data, the accuracy of each path prediction result is calculated, and the optimal feature combination with high accuracy is selected.
[0025] This invention provides a machine learning-based telephone lead generation system, comprising the following modules:
[0026] Sample Feature Library Construction Module: Constructs a sample feature library including both harmful and normal phone numbers;
[0027] Data preprocessing module: preprocesses the feature data in the sample feature library and divides it into feature training set and feature test set;
[0028] Decision tree construction module: Constructs decision trees based on the feature training set;
[0029] Decision rule filtering module: Based on the feature test set, the decision rules in the decision tree are filtered to obtain the feature combination that performs best on the test set and the segmentation threshold corresponding to the feature combination;
[0030] Call number identification module: Collects new batches of call numbers and obtains the feature data of the call numbers. Based on the optimal feature combination and the segmentation threshold corresponding to the feature combination, it identifies the feature data of the call numbers and predicts whether the call number is a harmful number or a normal number.
[0031] Feature timed update module: Timed updates of the optimal feature combination and its corresponding segmentation threshold.
[0032] The present invention provides a computer-readable storage medium for storing one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform any of the above-described machine learning-based telephone lead generation methods.
[0033] Beneficial effects: The present invention has the following advantages: 1. The present invention uses machine learning algorithms to classify sample feature data and select the feature combination and segmentation threshold with the best traffic-driving effect, which can more accurately identify harmful calls and reduce false judgments and false misses; 2. By updating the optimal feature combination at regular intervals, it ensures that even when the characteristics of harmful numbers change, it can still achieve a good traffic-driving effect, so as to detect as many harmful calls as possible with as few calls as possible. Attached Figure Description
[0034] Figure 1 This is a schematic diagram of the method flow;
[0035] Figure 2 A schematic diagram of the apparatus for implementing the method. Detailed Implementation
[0036] The technical solution of the present invention will be described in detail below with reference to the embodiments and accompanying drawings.
[0037] like Figure 1 As shown, the machine learning-based telephone lead generation method of this invention specifically includes the following steps:
[0038] S1: Construct a sample feature library including both harmful and normal phone numbers;
[0039] The case system is connected through system interface or file synchronization. The caller's number and called number of harmful calls are continuously extracted from the case system as harmful number samples and normal number samples, respectively, to dynamically build a sample feature library. The sample feature library includes basic features based on communication behavior and statistical features obtained by analyzing the basic features.
[0040] Basic features refer to characteristic dimensions in user calls that do not require statistical calculation, such as caller ID, caller's location, called ID, called location, call start time, ring start time, answer time, and call end time. Statistical features refer to results obtained by analyzing and calculating based on user communication behavior, such as caller ID connection rate, called ID connection rate, caller ID dispersion, called ID dispersion, and the total number of days within 30 days when more than 50 calls were made per day.
[0041] The sample feature library can automatically construct data features for corresponding dimensions through well-written script code programs, providing training and analysis data for subsequent steps.
[0042] S2: Preprocess the feature data in the sample feature library and divide it into feature training set and feature test set. Preprocessing includes data cleaning, missing value handling, etc., to ensure data quality.
[0043] S3: Construct a decision tree based on the feature training set;
[0044] A decision tree consists of a root node, internal nodes, leaf nodes, and branches. The root node and internal nodes are essentially splitting features. The root node is the starting point of the decision tree and typically selects the most discriminative feature. Internal nodes are nodes that split from the root node; each internal node selects a feature and a threshold to split into two child nodes, corresponding to the subsets of data that satisfy and do not satisfy the threshold condition, respectively. Leaf nodes represent the categories.
[0045] The decision tree construction process specifically includes:
[0046] S3.1: Feature selection: Select the most discriminative feature from the feature training set as the root node or internal node;
[0047] The most discriminative feature is typically the one that most easily distinguishes different categories. For each splitting feature, we aim for its branches to contain data belonging to the same category as much as possible, i.e., increasing "purity." "Purity" is usually measured using the Gini index, information gain, or information gain ratio. In this embodiment, the feature with the lowest Gini index is selected as the most discriminative feature. For example, based on the preprocessed feature training set F, the Gini index of each feature is first calculated, and the feature with the lowest Gini index is selected as the root node. The formula for calculating the Gini index for a node is:
[0048]
[0049] Where m is the number of samples at a node, n is the number of categories, and c is the number of samples in category i. i .
[0050] S3.2: Threshold selection: Determine a segmentation threshold for the features selected in S3.1;
[0051] The optimal segmentation threshold is determined by setting a segmentation threshold to minimize the impurity of child nodes, which is measured by the Gini index.
[0052] S3.3: Recursive Splitting: Starting from the root node, based on the selected features and segmentation threshold, the feature training set is split into several internal nodes. For each internal node, the most discriminative feature and corresponding segmentation threshold are selected from the remaining feature training set F for further splitting. This feature selection and threshold selection process is repeated until a stopping condition is met. For the child nodes after the split of each internal node, feature selection and threshold selection are further performed from the remaining feature training set F, and this process is repeated until a certain stopping condition is met.
[0053] Stopping conditions: The decision tree reaches a predetermined maximum depth, or the number of samples in a node is less than a certain threshold, or the purity of samples within a node reaches a certain standard (e.g., the Gini index of a node).
[0054] S3.4: Optimize decision tree pruning, i.e. remove some branches to prevent overfitting. The criteria for removal include reaching the predetermined maximum tree depth, the number of samples in a node being less than a certain threshold, and the purity of samples within a node reaching a certain standard.
[0055] The final decision tree consists of a series of paths from the root node to the leaf nodes. Each path represents a decision rule, and each decision rule contains multiple feature combinations. Each feature combination has its corresponding segmentation threshold, and these combinations collectively determine the classification result of the data along that path. Specifically:
[0056] Root node: The starting point of the decision tree, usually selected as the most discriminative feature.
[0057] Internal nodes: Each internal node represents a feature and a threshold, used to further split the dataset.
[0058] Leaf node: The endpoint of the decision tree, representing the final classification result.
[0059] S4: Based on the feature test set, filter the decision rules in the decision tree to obtain the feature combination that performs best on the test set and the segmentation threshold corresponding to the feature combination;
[0060] The selection process is as follows: For each test data point, starting from the root node of the decision tree, a path is selected based on the features until a leaf node is reached. The leaf node provides the final classification prediction result, such as normal or harmful. Based on the actual labels of the test data, the accuracy of the prediction results for each path is calculated, and the optimal feature combination with the highest accuracy is selected.
[0061] Accuracy is the proportion of correctly predicted samples out of the total number of samples. If the accuracy calculated by a decision rule exceeds a set threshold, the decision rule is considered to have "high accuracy". To further improve the reliability of the screening, methods such as cross-validation can be used to select the decision rule that performs best on multiple test sets.
[0062] S5: Collect the call number and obtain its feature data, including basic features and statistical features. Based on the optimal feature combination and the segmentation threshold corresponding to the feature combination, identify the feature data of the call number to predict whether the call number is a harmful number or a normal number.
[0063] S6: Repeat steps S1-S5 periodically to update the optimal feature combination and its corresponding segmentation threshold, ensuring that even when the characteristics of harmful numbers change, a good traffic diversion effect can still be achieved, so as to detect as many harmful calls as possible with as few calls as possible.
[0064] like Figure 2 As shown, the present invention further provides a machine learning-based telephone lead generation device, including a processor, a bus, a storage medium, an internal memory, a display screen, a network interface, and an input device. The memory stores an executable computer program, and when the processor executes the computer program through the system bus, it is used to implement the above-mentioned machine learning-based telephone lead generation method.
Claims
1. A telephone lead generation method based on machine learning, characterized in that, Includes the following steps: S1: Extract the calling and called numbers of harmful calls from the case system as harmful number samples and normal number samples, respectively, and construct a sample feature library; The sample feature library includes basic features based on communication behavior and statistical features obtained by analyzing the basic features; S2: Preprocess the feature data in the sample feature library and divide it into a feature training set and a feature test set; S3: Construct a decision tree based on the feature training set; the generated decision tree contains a series of paths from the root node to the leaf node, each path represents a decision rule, each decision rule contains multiple feature combinations, and each feature combination has a corresponding segmentation threshold; the construction process includes: S3.1: Feature selection: Select the most discriminative feature from the feature training set as the root node or internal node; S3.2: Threshold selection: Determine a segmentation threshold for the most discriminative feature selected; S3.3: Recursive splitting. Starting from the root node, based on the selected features and segmentation threshold, the feature training set is split into several internal nodes. For each internal node, the most discriminative feature and corresponding segmentation threshold are selected from the remaining feature training set for further splitting. This feature selection and threshold selection process is repeated until the stopping condition is met. For the child nodes after the split of each internal node, feature selection and threshold selection are further performed from the remaining feature training set, and this process is repeated until the stopping condition is met. S3.4: Optimize decision tree pruning; Among them, the feature with the lowest Gini index was selected as the most distinctive feature; The formula for calculating the Gini index is: ， Where m is the number of samples at a node, n is the number of categories, and c is the number of samples in category i. i ; The stopping conditions are reaching a predetermined maximum tree depth, or the number of node samples is less than a certain set threshold, or the purity of samples within a node reaches a certain set standard. S4: Based on the feature test set, filter the decision rules in the decision tree to obtain the feature combination that performs best on the test set and the segmentation threshold corresponding to the feature combination; The optimal feature combination is selected by using a cross-validation screening method, which selects the best-performing feature combination on multiple test sets. Each test data starts from the root node of the decision tree, and the path is selected according to the features of the test data until a leaf node is reached. The leaf node gives the final classification prediction result, such as normal or harmful. Based on the actual label of the test data, the accuracy of each path prediction result is calculated, and the optimal feature combination with high accuracy is selected. S5: Collect new batches of call numbers and obtain the feature data of the call numbers. Based on the optimal feature combination and the segmentation threshold corresponding to the feature combination, identify the feature data of the call numbers to predict whether the call number is a harmful number or a normal number. S6: Repeat steps S1-S5 to periodically update the optimal feature combination and its corresponding segmentation threshold.
2. The telephone lead generation method based on machine learning according to claim 1, characterized in that, In S1, case system integration is completed through system interfaces or file synchronization.
3. The telephone lead generation method based on machine learning according to claim 1, characterized in that, The preprocessing methods in S2 include data cleaning and missing value handling.
4. A system applied to the machine learning-based telephone lead generation method as described in claim 1, characterized in that, Includes the following modules: Sample Feature Library Construction Module: Constructs a sample feature library including both harmful and normal phone numbers; Data preprocessing module: preprocesses the feature data in the sample feature library and divides it into feature training set and feature test set; Decision tree construction module: Constructs decision trees based on the feature training set; Decision rule filtering module: Based on the feature test set, the decision rules in the decision tree are filtered to obtain the feature combination that performs best on the test set and the segmentation threshold corresponding to the feature combination; Call number identification module: Collects new batches of call numbers and obtains the feature data of the call numbers. Based on the optimal feature combination and the segmentation threshold corresponding to the feature combination, it identifies the feature data of the call numbers and predicts whether the call number is a harmful number or a normal number. Feature timed update module: Timed updates of the optimal feature combination and its corresponding segmentation threshold.
5. A computer-readable storage medium for storing one or more programs, characterized in that, The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any of the methods according to claims 1 to 3.