Outbound contact method and device based on customer psychological prediction, equipment and medium

Abstract

The application provides a customer psychological prediction-based outbound contact method and device, equipment and a medium, comprising: collecting daily outbound full data of a business operator, and constructing a customer call portrait based on the collected daily outbound full data of the business operator; predicting a customer rejection psychology level by using an XGBoost multi-classification prediction model based on the customer call portrait; dynamically switching an outbound calling number type according to a customer rejection psychology level determination result, configuring a business identifier for a switched virtual number, and performing an outbound operation by using a corresponding calling number.

Description

Technical Field

[0001] This invention relates to the field of logistics management technology, and more specifically, to outbound calling methods, devices, equipment, and media based on customer psychological prediction. Background Technology

[0002] In outbound calling services, especially in logistics fulfillment scenarios, sales personnel need to proactively call customers to verify information such as logistics receipt and package delivery. Current outbound calling models often rely on a fixed set of official carrier numbers. Over time, these official numbers are easily flagged by customers as marketing or harassing calls, leading to a continuously increasing rejection rate. Furthermore, existing outbound calling strategies fail to analyze customers' historical calling behavior to assess their level of rejection and cannot differentiate caller ID settings. Highly resistant customers continue to use official numbers, resulting in low connection rates and poor utilization of outbound calling resources.

[0003] Currently, most intelligent outbound calling solutions rely solely on call frequency to determine customer willingness to answer, resulting in a single-dimensional assessment that fails to incorporate multi-dimensional call feature modeling and analysis, such as historical call connection data, hang-up behavior, and call intervals. Conventional machine learning models cannot accurately capture the tiered psychology of customer call rejection, and the precision of multi-level classification is insufficient. Furthermore, existing solutions do not differentiate between valid and invalid calls in the statistical analysis of customer call rejection behavior, leading to inconsistent statistical methods for customer call rejection samples and further reducing the accuracy of predicting customer call rejection levels. Summary of the Invention

[0004] The main objective of this invention is to solve the technical problems of existing technologies, such as the inability to accurately fit the psychology of customer segmentation and rejection, and the insufficient accuracy of multi-level classification.

[0005] The first aspect of this invention provides an outbound calling method based on customer psychological prediction, comprising: Collect all daily outbound call data of salespersons, and build customer call profiles based on the collected daily outbound call data; Based on the customer call profile, the XGBoost multi-class prediction model is used to predict the customer's psychological level of rejection. Based on the customer's psychological level of rejection, the outbound caller ID type is dynamically switched, and a service identifier is configured for the switched virtual number. The corresponding caller ID is then used to perform the outbound call.

[0006] Optionally, in a first implementation of the first aspect of the present invention, the step of collecting all daily outbound call data of salespersons and constructing a customer call profile based on the collected daily outbound call data of salespersons includes: Collect all daily outbound call data from sales representatives, including: waybill number, customer mobile phone number, call behavior, and call result data; wherein, the call behavior includes the duration of a single call; the call result data includes: unanswered call data, answered call data with valid call duration, and answered call data with invalid call duration, and answered call data with invalid call duration is regarded as unanswered call data. Link waybill numbers, customer mobile phone numbers, call behavior, and call result data together; Based on the business data of the associated customers within a preset time period, call features are extracted, including total call frequency, success rate of official number segment, cumulative number of rejections of official number segment, single call hang-up point, answering time preference, and call interval duration associated with waybill. Generate dynamically updated customer call profiles based on extracted features.

[0007] Optionally, in a second implementation of the first aspect of the present invention, the XGBoost multi-class prediction model includes: an input layer, a preprocessing layer, a multi-CART decision tree ensemble layer, a probability output layer, and a ranking discrimination layer; The input layer is used to generate a feature vector based on the combination of call features in the customer call profile. The preprocessing layer is used to preprocess the feature vector to obtain the preprocessed feature vector; The multi-CART decision tree ensemble layer is used to obtain a comprehensive prediction score based on the preprocessed feature vectors. The probability output layer is used to map the comprehensive prediction score to a standardized rejection probability value. The level discrimination layer is used to automatically classify and output the customer resistance level based on the standardized rejection probability value and a preset probability threshold range.

[0008] Optionally, in a third implementation of the first aspect of the present invention, the step of predicting the customer's psychological level of rejection using an XGBoost multi-class prediction model based on the customer call profile includes: The customer call profile, which includes features such as total call frequency, success rate of official number segment, cumulative number of rejections of official number segment, single call hang-up point, answering time preference, and call interval duration associated with waybill, is fed into the input layer to generate a feature vector. The feature vector is input into the preprocessing layer, which performs standardization and outlier removal on the input feature vector to obtain the preprocessed feature vector. The preprocessed feature vectors are input into the multi-CART decision tree ensemble layer, and multiple independent classification and regression trees are called to perform parallel operations to output rejection tendency scores respectively. The branch operation results of all trees are then superimposed to obtain the comprehensive prediction score. The comprehensive prediction score is input into the probability output layer, and the comprehensive prediction score is mapped to a standardized rejection probability value through the Softmax activation function. The standardized rejection probability value is input into the level discrimination layer, and the customer resistance level is automatically divided and output according to the preset probability threshold range.

[0009] Optionally, in a fourth implementation of the first aspect of the present invention, before using the XGBoost multi-class prediction model to predict the customer's psychological level of rejection, the XGBoost multi-class prediction model is trained, including: A historical outbound call sample dataset with real customer resistance level labels is constructed. The call features corresponding to the sample dataset are extracted and then processed through the input layer and preprocessing layer of the XGBoost multi-class prediction model before being sent to the multi-CART decision tree ensemble layer. The multi-CART decision tree ensemble layer employs a residual iterative training mechanism. During the iteration process, individual classification and regression trees are generated sequentially. Each individual classification and regression tree branches layer by layer according to feature splitting rules, outputting an initial rejection tendency score. The XGBoost multi-class prediction model aims to minimize the loss function. It superimposes this initial rejection tendency score onto the historical cumulative prediction scores to obtain the current comprehensive prediction score. This score is then mapped through the probability output layer to obtain the prediction probability. The loss value is calculated by combining this with the sample's true label, and the weights of the currently generated individual classification and regression tree are optimized in reverse to fit the sample prediction residuals. This iterative process is repeated to generate multiple classification and regression trees until the loss function converges, at which point the iteration terminates, completing the training of the XGBoost multi-class prediction model.

[0010] Optionally, in the fifth implementation of the first aspect of the present invention, the step of dynamically switching the outbound calling number type based on the customer's psychological rejection level determination result, configuring a service identifier for the switched virtual number, and performing outbound calls using the corresponding calling number includes: When the customer's psychological rejection level meets the preset requirements, the built-in 11-digit personal virtual number pool is retrieved, and an available virtual number is randomly assigned as the calling number for this outbound call. At the same time, through the operator's number identification technology, the business label is displayed on the outbound call interface of the 11-digit virtual mobile number.

[0011] Optionally, in a sixth implementation of the first aspect of the present invention, the method further includes: real-time monitoring of the remaining available quota in the virtual number pool; when the remaining available quota in the virtual number pool is less than a preset value, switching back to the official number segment to complete the call operation.

[0012] A second aspect of the present invention provides an outbound calling device based on customer psychological prediction, comprising: The customer call profile building module is used to collect all outbound call data of salespersons in their daily work and build customer call profiles based on the collected data. The prediction module is used to predict the customer's psychological level of rejection based on the customer call profile using the XGBoost multi-class prediction model. The outbound call switching module is used to dynamically switch the outbound caller ID type based on the customer's psychological rejection level, configure a service identifier for the switched virtual number, and execute the outbound call operation using the corresponding caller ID.

[0013] Optionally, in a first implementation of the second aspect of the present invention, the customer call profile construction module includes: The data collection submodule is used to collect all outbound business data of salespersons every day, including: waybill number, customer mobile phone number, call behavior and call result data; wherein, the call behavior includes the single call time; the call result data includes: unanswered call data, answered call data with valid call duration and answered call data with invalid call duration, and answered call data with invalid call duration is regarded as unanswered call data; The data interconnection submodule is used to interconnect waybill numbers, customer mobile phone numbers, call behavior, and call result data; The call feature extraction submodule is used to extract call features based on the business data of associated customers within a preset time period, including total call frequency, official number segment connection success rate, official number segment cumulative number of rejections, single call hang-up node, answering time preference, and waybill associated call interval duration. The profile building submodule is used to generate dynamically updatable customer call profiles based on extracted features.

[0014] Optionally, in a second implementation of the second aspect of the present invention, the XGBoost multi-class prediction model includes: an input layer, a preprocessing layer, a multi-CART decision tree ensemble layer, a probability output layer, and a ranking discrimination layer; The input layer is used to generate a feature vector based on the combination of call features in the customer call profile. The preprocessing layer is used to preprocess the feature vector to obtain the preprocessed feature vector; The multi-CART decision tree ensemble layer is used to obtain a comprehensive prediction score based on the preprocessed feature vectors. The probability output layer is used to map the comprehensive prediction score to a standardized rejection probability value. The level discrimination layer is used to automatically classify and output the customer resistance level based on the standardized rejection probability value and a preset probability threshold range.

[0015] Optionally, in a third implementation of the second aspect of the present invention, the prediction module includes: The feature vector generation submodule is used to send the customer call profile, which includes features such as total call frequency, official number segment connection success rate, official number segment cumulative rejection count, single call hang-up node, answering time preference, and waybill associated call interval duration, into the input layer to generate feature vectors. The preprocessing submodule is used to input the feature vector into the preprocessing layer, and the preprocessing layer performs standardization and outlier removal on the input feature vector to obtain the preprocessed feature vector. The comprehensive prediction score generation submodule is used to input the preprocessed feature vector into the multi-CART decision tree ensemble layer, call multiple independent classification and regression trees to perform parallel operations and output rejection tendency scores respectively, and then superimpose the branch operation results of all trees to obtain the comprehensive prediction score; The probability value generation submodule is used to input the comprehensive prediction score into the probability output layer and map the comprehensive prediction score to a standardized rejection probability value through the Softmax activation function. The rejection psychological level generation submodule is used to input standardized rejection probability values ​​into the level discrimination layer, and automatically classify and output customer resistance levels according to preset probability threshold ranges.

[0016] Optionally, in a fourth implementation of the second aspect of the present invention, the XGBoost multi-class prediction model is trained before predicting the customer's psychological level of rejection using the XGBoost multi-class prediction model, including: A historical outbound call sample dataset with real customer resistance level labels is constructed. The call features corresponding to the sample dataset are extracted and then processed through the input layer and preprocessing layer of the XGBoost multi-class prediction model before being sent to the multi-CART decision tree ensemble layer. The multi-CART decision tree ensemble layer employs a residual iterative training mechanism. During the iteration process, individual classification and regression trees are generated sequentially. Each individual classification and regression tree branches layer by layer according to feature splitting rules, outputting an initial rejection tendency score. The XGBoost multi-class prediction model aims to minimize the loss function. It superimposes this initial rejection tendency score onto the historical cumulative prediction scores to obtain the current comprehensive prediction score. This score is then mapped through the probability output layer to obtain the prediction probability. The loss value is calculated by combining this with the sample's true label, and the weights of the currently generated individual classification and regression tree are optimized in reverse to fit the sample prediction residuals. This iterative process is repeated to generate multiple classification and regression trees until the loss function converges, at which point the iteration terminates, completing the training of the XGBoost multi-class prediction model.

[0017] Optionally, in a fifth implementation of the second aspect of the present invention, the outbound call reach switching module includes: When the customer's psychological rejection level meets the preset requirements, the built-in 11-digit personal virtual number pool is retrieved, and an available virtual number is randomly assigned as the calling number for this outbound call. At the same time, through the operator's number identification technology, the business label is displayed on the outbound call interface of the 11-digit virtual mobile number.

[0018] Optionally, in a sixth implementation of the second aspect of the present invention, the device further includes: a real-time monitoring module, used to monitor the remaining available quota of the virtual number pool in real time, and when the remaining available quota in the virtual number pool is less than a preset value, to switch back to the official number segment to complete the call operation.

[0019] A third aspect of the present invention provides an electronic device, the electronic device comprising a memory and at least one processor, wherein the memory stores instructions; The at least one processor invokes the instructions in the memory to cause the electronic device to perform the various steps of the outbound call outreach method based on customer psychological prediction as described above.

[0020] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the steps of the outbound call outreach method based on customer psychological prediction as described above.

[0021] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention collects all outbound business data of logistics, distinguishes between valid call duration and invalid call duration, and classifies invalid calls as unanswered calls, thus unifying the statistical criteria for customer rejection behavior. At the same time, it extracts multi-dimensional exclusive call features such as total call frequency, official number segment connection success rate, and waybill-related call interval to construct a dynamic customer call profile, comprehensively depicting customers' historical answering, hanging up, and outbound call tolerance behaviors, and accurately reflecting customers' outbound call resistance psychology. 2. The XGBoost multi-class prediction model of this invention relies on the input layer, preprocessing layer, multi-CART decision tree integration layer, probability output layer, and level discrimination layer to complete hierarchical data processing; in the online prediction stage, multiple CART decision trees with fixed parameters are called for parallel scoring calculation, and the Softmax activation function is combined to complete multi-level probability mapping, efficiently and hierarchically determine the customer's psychological level of rejection, and adapt to the needs of fine classification of low, medium and high levels of customer resistance. Compared with traditional single machine learning models, the customer rejection level prediction accuracy is higher and the adaptability is stronger. 3. This invention dynamically switches the outbound caller ID type based on the customer's psychological level of refusal. For highly resistant customers, it adapts to 11-digit personal virtual numbers for outbound calls and attaches business tags to complete the caller ID display, weakening the harassment label attribute of official number segments and effectively improving the outbound call connection rate for highly resistant customers. At the same time, it is equipped with a virtual number pool surplus monitoring backup mechanism, which automatically switches back to official number segments for outbound calls when number resources are insufficient, avoiding the risk of outbound call operation interruption and ensuring the continuous and stable operation of outbound call services. Attached Figure Description

[0022] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a first flowchart of an outbound call outreach method based on customer psychological prediction provided in an embodiment of the present invention.

[0023] Figure 2 This is a second flowchart of the outbound call outreach method based on customer psychological prediction provided in an embodiment of the present invention.

[0024] Figure 3 This is a schematic diagram of an outbound call reaching device based on customer psychological prediction provided in an embodiment of the present invention.

[0025] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0026] This invention provides a method, apparatus, device, and medium for outbound calling based on customer psychological prediction, comprising: collecting all daily outbound call data of salespersons; constructing a customer call profile based on the collected daily outbound call data; predicting the customer's rejection psychological level using an XGBoost multi-classification prediction model based on the customer call profile; dynamically switching the outbound caller ID type according to the customer rejection psychological level determination result, configuring a business identifier for the switched virtual number, and performing outbound calls using the corresponding caller ID. This invention solves the technical problems of existing technologies that cannot accurately fit the stratified rejection psychology of customers and have insufficient accuracy in multi-classification level division.

[0027] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0028] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 The first embodiment of the outbound call outreach method based on customer psychological prediction in this invention includes: 101. Collect all daily outbound call data of salespersons and build customer call profiles based on the collected daily outbound call data; In this embodiment, all outbound business data of the salesperson are collected daily, including: waybill number, customer mobile phone number, call behavior and call result data; The calling behavior includes the duration of a single call; The call result data includes: unanswered call data, answered call data with valid call duration, and answered call data with invalid call duration, and answered call data with invalid call duration is regarded as unanswered call data; The collected waybill numbers, customer identity mobile phone numbers, outbound call behavior data, and call result data are bound and associated one by one to establish a four-dimensional association database of "customer-waybill-call behavior-call result".

[0029] Based on the business data of the associated customers within a preset time period, call features are extracted, including total call frequency, success rate of official number segment, cumulative number of rejections of official number segment, single call hang-up point, answering time preference, and call interval duration associated with waybill. The total call frequency refers to the cumulative number of outbound calls made to the customer within a preset time period, reflecting the intensity of calls made to the customer. The higher the total number of outbound calls within the preset period, the more frequently the customer is disturbed, which can easily lead to annoyance and resistance, and significantly increase the probability of subsequent outbound call rejection. The lower the outbound call frequency, the less fatigued the customer is, and the stronger their willingness to answer. The success rate of connecting to the official number segment is the ratio of the number of valid calls to the total number of calls when the official fixed service number segment calls the customer. The lower the success rate, the more likely the customer has a long-term aversion, distrust, and resistance to the official service number, which is a fixed negative emotion. The higher the success rate, the more accepting the customer is of the official number and the weaker the refusal mentality. The cumulative number of rejected calls from official numbers, and the total number of missed or invalid calls made to the customer from official fixed service numbers, directly reflect the customer's level of resistance to official service numbers. The more rejected calls there are, the more likely the customer is to actively refuse official calls, with resistance accumulating and intensifying, eventually leading to habitual rejection. The fewer the rejections, the less negative emotions accumulate.

[0030] The single call hang-up point is determined by statistically analyzing the time points at which customers hang up after answering a call. These are categorized into levels such as hang-up immediately upon connection, hang-up within 10 seconds of communication, and hang-up within 30 seconds of communication, reflecting the intensity of customer resistance. Hang-ups immediately upon connection or quickly within 10 seconds indicate extremely strong customer resistance after recognizing the outbound call and an unwillingness to communicate. Hang-ups after 30 seconds or more suggest weaker customer resistance, indicating only a brief communication session. The earlier the hang-up, the stronger the immediate refusal to answer.

[0031] The call answering time preference is calculated by analyzing the time periods when customers frequently answer and reject calls, distinguishing between peak hours in the morning, noon, evening, and late at night, and identifying the best contact window for customers. During peak call rejection times, such as the morning rush hour or late at night, customers are more likely to temporarily reject calls due to busyness or being disturbed. During peak call answering times, customers are more relaxed and have less resistance.

[0032] The call intervals associated with waybills are statistically analyzed, tracking the time intervals between key milestones such as waybill generation, update, and completion, and outbound calls. This reflects differences in customer attitudes towards different business stages. Calls made immediately after waybill generation / update elicit high customer interest and low resistance; however, frequent calls long after waybill completion are perceived as pointless harassment and a sign of unsolicited calls, significantly increasing the likelihood of rejection. Inappropriate call intervals are a core factor influencing call rejections in various business scenarios.

[0033] Generate dynamically updated customer call profiles based on extracted features.

[0034] The customer call profile has six key features, including static historical habits (frequency, number of rejections, and connection rate), dynamic emotional intensity (hang-up point), spatiotemporal scenarios (time period preference), and business scenarios (call interval), which comprehensively cover the factors influencing customer rejection psychology and complement and correct each other.

[0035] This embodiment uses refined data classification rules to filter out invalid data interference, avoid false call answering data from affecting the accuracy of the profile, and ensure the authenticity and validity of the underlying data source.

[0036] 102. Based on the customer call profile, use the XGBoost multi-class prediction model to predict the customer's psychological level of rejection. In this embodiment, the XGBoost multi-class prediction model includes: an input layer, a preprocessing layer, a multi-CART decision tree ensemble layer, a probability output layer, and a ranking discrimination layer; The input layer is used to generate a feature vector based on the combination of call features in the customer call profile. The preprocessing layer is used to preprocess the feature vector to obtain the preprocessed feature vector; The multi-CART decision tree ensemble layer is used to obtain a comprehensive prediction score based on the preprocessed feature vectors. The probability output layer is used to map the comprehensive prediction score to a standardized rejection probability value. The level discrimination layer is used to automatically classify and output the customer resistance level based on the standardized rejection probability value and a preset probability threshold range.

[0037] Before using the XGBoost multi-class prediction model to predict the psychological level of customer rejection, the XGBoost multi-class prediction model is trained, including: A historical outbound call sample dataset with real customer resistance level labels is constructed. The call features corresponding to the sample dataset are extracted and then processed through the input layer and preprocessing layer of the XGBoost multi-class prediction model before being sent to the multi-CART decision tree ensemble layer. The multi-CART decision tree ensemble layer employs a residual iterative training mechanism. During the iteration process, individual classification and regression trees are generated sequentially. Each individual classification and regression tree branches layer by layer according to feature splitting rules, outputting an initial rejection tendency score. The XGBoost multi-class prediction model aims to minimize the loss function. It superimposes this initial rejection tendency score onto the historical cumulative prediction scores to obtain the current comprehensive prediction score. This score is then mapped through the probability output layer to obtain the prediction probability. The loss value is calculated by combining this with the sample's true label, and the weights of the currently generated individual classification and regression tree are optimized in reverse to fit the sample prediction residuals. This iterative process is repeated to generate multiple classification and regression trees until the loss function converges, at which point the iteration terminates, completing the training of the XGBoost multi-class prediction model.

[0038] This embodiment utilizes a residual iterative optimization mechanism to correct prediction errors one by one, avoiding overfitting and underfitting issues, and significantly improving the model's accuracy in recognizing customer resistance. The trained model can accurately adapt to real-world outbound calling scenarios and, compared to traditional machine learning models, can precisely distinguish different levels of customer resistance.

[0039] The step of predicting the customer's psychological level of rejection based on the customer call profile using the XGBoost multi-class prediction model includes: The customer call profile, which includes features such as total call frequency, success rate of official number segment, cumulative number of rejections of official number segment, single call hang-up point, answering time preference, and call interval duration associated with waybill, is fed into the input layer to generate a feature vector. The feature vector is input into the preprocessing layer, which performs standardization and outlier removal on the input feature vector to obtain the preprocessed feature vector; thus completely solving the problems of inconsistent dimensions of multiple features and interference from outlier data. The preprocessed feature vectors are input into the multi-CART decision tree ensemble layer, and multiple independent classification and regression trees are called to perform parallel operations to output rejection tendency scores respectively. The branch operation results of all trees are then superimposed to obtain the comprehensive prediction score. The comprehensive prediction score is input into the probability output layer. A Softmax activation function is used to perform a non-linear transformation on the score, mapping the undefined comprehensive prediction score to a standardized rejection probability value between 0 and 1. The closer the probability value is to 1, the stronger the customer's tendency to reject outbound calls; the closer it is to 0, the higher the customer's willingness to answer.

[0040] The standardized rejection probability value is input into the level discrimination layer, and the customer resistance level is automatically divided and output according to the preset probability threshold range.

[0041] More specifically, the rating system incorporates a three-level probability threshold range to standardize the classification of customer rejection psychological levels. Specific rating rules include: Low resistance level: rejection rate 0-0.3%, customers have a strong willingness to answer and a high acceptance of official outbound calls; Medium level of resistance: probability of rejection 0.3-0.7, customers are generally not willing to answer, and there is a certain probability of rejection; High resistance level: rejection rate 0.7-1.0, customers have a strong desire to refuse, and calls from official numbers are very likely to be hung up or rejected.

[0042] The XGBoost multi-class prediction model learns the non-linear relationship between each feature and the rejection probability to accurately quantify the overall level of customer resistance, ultimately supporting the subsequent differentiated outbound calling strategy for number switching.

[0043] 103. Based on the customer's psychological level of rejection, dynamically switch the outbound caller ID type, configure a service identifier for the switched virtual number, and use the corresponding caller ID to perform outbound calls.

[0044] In this embodiment, when the customer's psychological resistance level is high, the built-in 11-digit personal virtual number pool is retrieved. The numbers are all virtual numbers from compliant operators and look exactly the same as ordinary personal mobile phone numbers, thus avoiding the customer's resistance to official fixed number segments. After receiving the high resistance level judgment result, the system automatically selects idle virtual numbers with no call occupation and no risk control risk from the virtual number pool as the outbound calling number. By using operator-specific number identification technology, standardized business labels such as "Logistics Notification", "Business Follow-up", and "After-sales Consultation" are displayed simultaneously on the incoming call interface of mobile phones that make outbound calls from virtual numbers. This retains the low-resistance advantage of personal numbers while solving the problem of unknown virtual numbers being mistaken for spam calls. The switched virtual number with service identifier was used to complete this customer outbound call operation.

[0045] This embodiment breaks through the bottleneck of traditional outbound calls to fixed numbers and indiscriminate reach. Through an intelligent closed loop of "data profiling + psychological prediction + dynamic reach", it accurately matches the customer's psychology of answering the call, effectively reduces the customer's willingness to refuse, and greatly improves the success rate of outbound call reach, while ensuring the compliance, stability and efficiency of outbound call business.

[0046] Please see Figure 2 The second embodiment of the outbound call outreach method based on customer psychological prediction in this invention includes: 201. Collect all daily outbound call data of salespersons and build customer call profiles based on the collected daily outbound call data; 202. Based on the customer call profile, use the XGBoost multi-class prediction model to predict the customer's psychological level of rejection. 203. Based on the customer's psychological level of rejection, dynamically switch the outbound caller ID type, configure a service identifier for the switched virtual number, and use the corresponding caller ID to perform outbound calls; 204. Monitor the remaining available quota of the virtual number pool in real time. When the remaining available quota in the virtual number pool is less than the preset value, that is, when the remaining available quota in the virtual number pool is insufficient, switch back to the official number segment to complete the call operation and ensure that the outbound call service is not interrupted.

[0047] The outbound calling method based on customer psychological prediction in the embodiments of the present invention has been described above. The outbound calling device based on customer psychological prediction in the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 3 One embodiment of the outbound call reach device based on customer psychological prediction in this invention includes: The customer call profile building module 301 is used to collect all daily outbound call data of salespersons and build customer call profiles based on the collected daily outbound call data of salespersons. In this embodiment, the customer call profile construction module 301 includes: The data acquisition submodule 3011 is used to collect all outbound business data of salespersons every day, including: waybill number, customer mobile phone number, call behavior and call result data; wherein, the call behavior includes the single call time; the call result data includes: unanswered call data, answered call data with valid call duration and answered call data with invalid call duration, and answered call data with invalid call duration is regarded as unanswered call data; The data interconnection submodule 3012 is used to interconnect waybill numbers, customer mobile phone numbers, call behavior and call result data to achieve the integration of outbound call data throughout the customer's entire life cycle, break down data silos, and fully restore the customer's overall behavior habits for business outbound calls, rather than the one-sided characteristics of a single call.

[0048] The call feature extraction submodule 3013 is used to extract call features based on the associated customer's business data within a preset time period, including total call frequency, official number segment connection success rate, official number segment cumulative rejection count, single call hang-up node, answering time period preference, and waybill associated call interval duration. The profile building submodule 3014 is used to generate dynamically updatable customer call profiles based on extracted features.

[0049] This embodiment integrates six core quantitative features to generate standardized, parsable, and iterative customer-specific call profiles. A dynamic update mechanism is also implemented, automatically synchronizing the latest outbound call data every 24 hours, updating customer characteristic parameters, and revising the profile content in real time. This customer call profile is real-time, dynamic, and accurate, able to iterate in sync with changes in customer answering habits and resistance, avoiding the problems of outdated and distorted static profiles.

[0050] Prediction module 302 is used to predict the customer's psychological level of rejection based on the customer call profile using the XGBoost multi-class prediction model. In this embodiment, the prediction module 302 includes: The feature vector generation submodule 3021 is used to send the customer call profile, which includes features such as total call frequency, official number segment connection success rate, official number segment cumulative rejection count, single call hang-up node, answering time period preference, and waybill associated call interval duration, into the input layer to generate feature vectors. The preprocessing submodule 3022 is used to input the feature vector into the preprocessing layer, and use the preprocessing layer to perform standardization processing and outlier removal processing on the input feature vector to obtain the preprocessed feature vector. The comprehensive prediction score generation submodule 3023 is used to input the preprocessed feature vector into the multi-CART decision tree integration layer, call multiple independent classification and regression trees to perform parallel operations and output rejection tendency scores respectively, and then superimpose the branch operation results of all trees to obtain the comprehensive prediction score; The probability value generation submodule 3024 is used to input the comprehensive prediction score into the probability output layer and map the comprehensive prediction score to the standardized rejection probability value through the Softmax activation function. The rejection psychological level generation submodule 3025 is used to input standardized rejection probability values ​​into the level discrimination layer, and automatically divide and output customer resistance levels according to preset probability threshold ranges.

[0051] The outbound call switching module 303 is used to dynamically switch the outbound caller number type based on the customer's psychological rejection level, configure a service identifier for the switched virtual number, and perform outbound calls using the corresponding caller number. In this embodiment, the outbound call reach switching module 303 includes: When the customer's psychological rejection level meets the preset requirements, the built-in 11-digit personal virtual number pool is retrieved, and an available virtual number is randomly assigned as the calling number for this outbound call. At the same time, through the operator's number identification technology, the business label is displayed on the outbound call interface of the 11-digit virtual mobile number.

[0052] This embodiment uses a combination of "personal virtual number + explicit business identifier" to doubly reduce customers' reluctance to answer calls, increase the answer rate of highly resistant customers, and at the same time ensure the compliance and business identification of outbound calling services.

[0053] The real-time monitoring module 304 is used to monitor the remaining available quota of the virtual number pool in real time. When the remaining available quota in the virtual number pool is less than the preset value, it switches back to the official number segment to complete the call operation, avoiding the problem of outbound call interruption and call failure caused by the exhaustion of virtual numbers, ensuring the continuous and stable operation of outbound call operations, eliminating the risk of business interruption, and improving the fault tolerance and stability of the outbound call system.

[0054] above Figure 3 The outbound call reaching device based on customer psychological prediction in the embodiments of the present invention will be described in detail from the perspective of modular functional entities. The electronic device in the embodiments of the present invention will be described in detail from the perspective of hardware processing.

[0055] Figure 4This is a schematic diagram of the structure of an electronic device 700 provided in an embodiment of the present invention. The electronic device 700 can vary significantly due to differences in configuration or performance, and may include one or more central processing units (CPUs) 710 (e.g., one or more processors) and a memory 720, and one or more storage media 730 (e.g., one or more mass storage devices) for storing application programs 733 or data 732. The memory 720 and storage media 730 can be temporary or persistent storage. The program stored in the storage media 730 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the electronic device 700. Furthermore, the processor 710 may be configured to communicate with the storage media 730 and execute the series of instruction operations in the storage media 730 on the electronic device 700.

[0056] Electronic device 700 may also include one or more power supplies 740, one or more wired or wireless network interfaces 750, one or more input / output interfaces 750, and / or one or more operating systems 731, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 4 The illustrated electronic device structure does not constitute a limitation on electronic devices and may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0057] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of an outbound call outreach method based on customer psychological prediction.

[0058] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0059] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0060] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for outbound calling based on customer psychological prediction, characterized in that, include: Collect all daily outbound call data of salespersons, and build customer call profiles based on the collected daily outbound call data; Based on the customer call profile, the XGBoost multi-class prediction model is used to predict the customer's psychological level of rejection. Based on the customer's psychological level of rejection, the outbound caller ID type is dynamically switched, and a service identifier is configured for the switched virtual number. The corresponding caller ID is then used to perform the outbound call.

2. The outbound calling method based on customer psychological prediction according to claim 1, characterized in that, The process involves collecting all daily outbound call data from sales representatives and constructing customer call profiles based on this data, including: Collect all daily outbound call data from sales representatives, including: waybill number, customer mobile phone number, call behavior, and call result data; wherein, the call behavior includes the duration of a single call; the call result data includes: unanswered call data, answered call data with valid call duration, and answered call data with invalid call duration, and answered call data with invalid call duration is regarded as unanswered call data. Link waybill numbers, customer mobile phone numbers, call behavior, and call result data together; Based on the business data of the associated customers within a preset time period, call features are extracted, including total call frequency, success rate of official number segment, cumulative number of rejections of official number segment, single call hang-up point, answering time preference, and call interval duration associated with waybill. Generate dynamically updated customer call profiles based on extracted features.

3. The outbound calling method based on customer psychological prediction according to claim 1, characterized in that, The XGBoost multi-class prediction model includes: an input layer, a preprocessing layer, a multi-CART decision tree ensemble layer, a probability output layer, and a ranking discrimination layer. The input layer is used to generate a feature vector based on the combination of call features in the customer call profile. The preprocessing layer is used to preprocess the feature vector to obtain the preprocessed feature vector; The multi-CART decision tree ensemble layer is used to obtain a comprehensive prediction score based on the preprocessed feature vectors. The probability output layer is used to map the comprehensive prediction score to a standardized rejection probability value. The level discrimination layer is used to automatically classify and output the customer resistance level based on the standardized rejection probability value and a preset probability threshold range.

4. The outbound calling method based on customer psychological prediction according to claim 3, characterized in that, The step of predicting the customer's psychological level of rejection based on the customer call profile using the XGBoost multi-class prediction model includes: The customer call profile, which includes features such as total call frequency, success rate of official number segment, cumulative number of rejections of official number segment, single call hang-up point, answering time preference, and call interval duration associated with waybill, is fed into the input layer to generate a feature vector. The feature vector is input into the preprocessing layer, which performs standardization and outlier removal on the input feature vector to obtain the preprocessed feature vector. The preprocessed feature vectors are input into the multi-CART decision tree ensemble layer, and multiple independent classification and regression trees are called to perform parallel operations to output rejection tendency scores respectively. The branch operation results of all trees are then superimposed to obtain the comprehensive prediction score. The comprehensive prediction score is input into the probability output layer, and the comprehensive prediction score is mapped to a standardized rejection probability value through the Softmax activation function. The standardized rejection probability value is input into the level discrimination layer, and the customer resistance level is automatically divided and output according to the preset probability threshold range.

5. The outbound calling method based on customer psychological prediction according to claim 4, characterized in that, Before using the XGBoost multi-class prediction model to predict the psychological level of customer rejection, the XGBoost multi-class prediction model is trained, including: A historical outbound call sample dataset with real customer resistance level labels is constructed. The call features corresponding to the sample dataset are extracted and then processed through the input layer and preprocessing layer of the XGBoost multi-class prediction model before being sent to the multi-CART decision tree ensemble layer. The multi-CART decision tree ensemble layer employs a residual iterative training mechanism. During the iteration process, individual classification and regression trees are generated sequentially. Each individual classification and regression tree branches layer by layer according to feature splitting rules, outputting an initial rejection tendency score. The XGBoost multi-class prediction model aims to minimize the loss function. It superimposes this initial rejection tendency score onto the historical cumulative prediction scores to obtain the current comprehensive prediction score. This score is then mapped through the probability output layer to obtain the prediction probability. The loss value is calculated by combining this with the sample's true label, and the weights of the currently generated individual classification and regression tree are optimized in reverse to fit the sample prediction residuals. This iterative process is repeated to generate multiple classification and regression trees until the loss function converges, at which point the iteration terminates, completing the training of the XGBoost multi-class prediction model.

6. The outbound calling method based on customer psychological prediction according to claim 1, characterized in that, The process of dynamically switching the outbound caller ID type based on the customer's level of rejection and configuring a service identifier for the switched virtual number, then using the corresponding caller ID to perform the outbound call, includes: When the customer's psychological rejection level meets the preset requirements, the built-in 11-digit personal virtual number pool is retrieved, and an available virtual number is randomly assigned as the calling number for this outbound call. At the same time, through the operator's number identification technology, the business label is displayed on the outbound call interface of the 11-digit virtual mobile number.

7. The outbound calling method based on customer psychological prediction according to claim 1, characterized in that, The method also includes: real-time monitoring of the remaining available quota in the virtual number pool; when the remaining available quota in the virtual number pool is less than a preset value, switching back to the official number segment to complete the call operation.

8. An outbound calling device based on customer psychological prediction, characterized in that, include: The customer call profile building module is used to collect all outbound call data of salespersons in their daily work and build customer call profiles based on the collected data. The prediction module is used to predict the customer's psychological level of rejection based on the customer call profile using the XGBoost multi-class prediction model. The outbound call switching module is used to dynamically switch the outbound caller ID type based on the customer's psychological rejection level, configure a service identifier for the switched virtual number, and execute the outbound call operation using the corresponding caller ID.

9. An electronic device comprising a memory and at least one processor, wherein the memory stores instructions; The at least one processor invokes the instructions in the memory to cause the electronic device to perform the steps of the outbound call outreach method based on customer psychological prediction as described in any one of claims 1-7.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the steps of the outbound call outreach method based on customer psychological prediction as described in any one of claims 1-7.

Images