Call recommendation method and system for outbound system

Abstract

The invention discloses a call recommendation method and system for an outbound system, and particularly relates to the technical field of intelligent recommendations, the outbound system obtains customer information and historical interaction records, and data preprocessing is performed to ensure the integrity and accuracy of data; dividing the customer data into sub-data sets according to different features, extracting features based on customer behavior data of each sub-data set, generating customer feature vectors, inputting the customer feature vectors into a pre-trained recommendation model, and generating recommendation indexes of customers; and finally, the recommendation results of all clients are merged and sorted to form a personalized outbound recommendation plan, so that the outbound efficiency and the client response rate are improved, and precise marketing is realized. According to the method, the personalized outbound content can be designed according to the emotional state and the behavior mode of the customer, so that the customer experience is improved, negative emotions caused by mismatched communication content are avoided, customer loss is further reduced, and customer loyalty is enhanced.

Description

Technical Field

[0001] The present invention relates to the field of intelligent recommendation technology, and more specifically, to a call recommendation method and system for an outbound call system. Background Art

[0002] In the modern market environment, outbound call systems, as an important tool in customer relationship management, are widely used in marketing, customer support, after-sales service and other fields. Traditional outbound call systems are usually based on relatively basic customer information (such as customer contact information, purchase records, etc.) to promote products or services to customers by making batch calls.

[0003] As enterprises pay more and more attention to customer relationship management, existing outbound call systems also face many challenges in technical application. Traditional outbound call systems usually use fixed rules or manual screening to select outbound call objects, and fail to make full use of the potential information in customer data. This method is not only time-consuming and labor-intensive, but also easily leads to invalid outbound calls, that is, frequent calls to customers who have no purchase intention or interest, which reduces the efficiency of outbound calls and may also cause customer disgust. In the prior art, outbound call systems mostly stay at the application level of basic information, such as simple customer classification or outbound calls in chronological order, lacking intelligent analysis methods. With the complexity of market demand and customer behavior, customers expect to receive personalized content that meets their needs, but traditional systems cannot effectively identify customers' behavioral preferences and response characteristics, resulting in a lack of pertinence in communication content. Customers often receive outbound call information whose content does not meet their needs, resulting in poor customer experience. This blind communication mode not only wastes resources, but may also lead to customer loss, and cannot truly improve customer response rate and sales conversion rate. Therefore, a call recommendation method and system for an outbound call system are proposed to solve the above problems. Summary of the invention

[0004] To achieve the above object, the present invention provides the following technical solutions: A call recommendation method for an outbound call system comprises the following steps: Obtain customer information and historical interaction records through the outbound call system, and pre-process the data to ensure its integrity and accuracy; Divide the acquired customer data into several sub-datasets, and classify each sub-dataset based on different customer characteristics; Extract features from the customer behavior data of each sub-dataset, input the extracted customer feature vector into the pre-trained recommendation model, and generate a recommendation index based on the customer's behavior characteristics; For each customer recommendation index, all customer recommendation results are merged and sorted, and ultimately a personalized outbound recommendation plan is generated.

[0005] In a preferred embodiment, basic customer information is extracted from a customer relationship management database and combined with customer historical interaction records extracted from a customer interaction information management database to form a preliminary customer data set, which is then preprocessed according to a preset preprocessing strategy, and the preprocessed customer data is stored in a comprehensive database and indexed.

[0006] In a preferred embodiment, the acquired customer data is divided into several sub-data sets, and each sub-data set is classified based on different customer characteristics, which means: According to the recommendation requirements of the outbound call system, the core feature dimensions of customer data are selected for segmentation. The core feature dimensions include customer response time, purchase behavior frequency, and call emotional state. Set classification standards for each feature dimension, including: The average response time of customers is divided into fast response customers, medium response customers and slow response customers according to a preset threshold; According to the customer's purchase records, customers are divided into high-frequency purchase customers, medium-frequency purchase customers and low-frequency purchase customers; Divide customers into high emotional response customers, neutral emotional response customers, and low emotional response customers based on their emotional scores; Based on the above feature classification standards, the customer data is divided into several sub-datasets, each of which corresponds to a specific single feature category or combined feature category, and a unique identifier is assigned to each divided sub-dataset to facilitate subsequent retrieval.

[0007] In a preferred embodiment, feature extraction of customer behavior data of each sub-data set refers to extracting customer response information, customer purchase information and customer call emotion information respectively, and then performing feature analysis operations to generate a customer response index, a customer purchase index and a customer emotion index respectively, and finally aggregating the customer response index, the customer purchase index and the customer emotion index together into a customer feature vector.

[0008] In a preferred embodiment, the logic for obtaining the customer response index is as follows: extract the time data of each response from the customer response information, and define the customer response time series as Indicates the timestamp of the i-th response, and n is the total number of responses; According to the customer's response time series, calculate the time interval between two adjacent responses to obtain the response time interval sequence Represents the time difference between the i-th and i+1-th responses; calculates the response time interval sequence The mean , reflecting the average response speed of customers; calculating the variance of the response time interval series , measures the consistency of customer responses; calculates the autocorrelation function of the response time interval sequence at the lag k moment , used to evaluate the periodicity and regularity of the response interval, the formula is: ; The customer response index is calculated based on the mean, variance and autocorrelation function of the response time interval, which is defined as: ; K represents the maximum time point of the autocorrelation function, which is used to balance the short-term and long-term response laws. are preset weighting coefficients, which are used to adjust the weights of mean, variance and autocorrelation function in the customer response index. is the maximum value of the variance, which is used to standardize the variance term, and R is the customer response index.

[0009] In a preferred embodiment, the logic for obtaining the customer purchase index is as follows: extract the time and amount data of each purchase from the customer purchase information, and define the customer purchase record sequence as , represents the timestamp of the i-th purchase, represents the corresponding purchase amount, m is the total number of purchase records; based on the customer's purchase record sequence, calculate the average frequency of the purchase time interval , reflects the customer's purchasing activity and calculates the average purchase amount , reflecting the average consumption level of customers and calculating the variance of purchase amount , used to measure the volatility of customer spending; the customer's purchase trend coefficient is calculated using the time-weighted average method to reflect the trend of customer spending habits over time. The calculation formula is: is the purchase trend coefficient, which indicates the influence of the customer's recent consumption habits on the purchase index. The customer purchase index is calculated based on the purchase frequency, the mean, variance and purchase trend coefficient of the purchase amount. The formula is as follows: is the maximum value of the purchase amount variance, which is used to standardize the variance. They are the influence coefficients of purchase frequency, average consumption, consumption stability and purchase trend, and D is the customer purchase index.

[0010] In a preferred embodiment, the logic for obtaining the customer sentiment index is: Extract the emotional state of each call from the customer's call emotional information and define the emotional state sequence as It represents the emotional state of the i-th call, with a value range of [0,1], indicating the emotional intensity from negative to positive, and v is the total number of emotional records; the maximum and minimum emotional state values ​​reflect the extreme bias of customer emotions, and the calculation formula is: It is an extreme bias value, which is used to quantify the extreme volatility of the customer's emotional state. The exponential function is used to perform nonlinear gain processing on the emotional change to enhance the impact of customer emotional fluctuations on the emotional index. The formula is: is the mood gain value; The customer sentiment index calculation formula is: ; is the preset attenuation factor, are all preset emotion infection coefficients, and F is the customer emotion index.

[0011] The recommendation model is obtained through convolutional neural network training. When used, the customer response index, customer purchase index and customer sentiment index are summarized as a customer feature vector and input into the recommendation model. The output result of the recommendation model is the recommendation index.

[0012] In a preferred embodiment, a call recommendation system for an outbound call system includes: The data acquisition and processing module obtains customer information and historical interaction records through the outbound call system, pre-processes the data to ensure the integrity and accuracy of the data, and stores the pre-processed customer data in the comprehensive database; The customer data segmentation module divides the customer data obtained from the comprehensive database into several sub-data sets, and each sub-data set is classified based on different customer characteristics; The feature extraction module extracts features from the customer behavior data of each sub-dataset, uses the extracted customer feature vector to input into the pre-trained recommendation model, and generates a recommendation index based on the customer's behavior characteristics; The outbound call recommendation plan generation module combines and sorts the recommendation results of all customers for each customer recommendation index, and finally generates a personalized outbound call recommendation plan.

[0013] Technical effects and advantages of the present invention: By analyzing multi-dimensional features such as customer response time, purchase behavior frequency, and call emotional state, the present invention can accurately identify high-potential customers and concentrate limited outbound call resources on customers who are more likely to respond, thereby significantly improving outbound call efficiency, avoiding repeated calls to low-willing customers by traditional systems, and reducing resource waste.

[0014] The system of the present invention generates a personalized outbound call recommendation plan through detailed analysis of customer characteristics. According to the customer's preferences and response patterns, the communication content and recommendation strategy are tailored for each customer, so that the customer can feel more targeted and considerate service, thereby improving customer satisfaction and reducing negative emotions caused by inconsistent communication content.

[0015] Traditional outbound call systems mainly rely on basic data, while the present invention integrates and analyzes multi-dimensional data such as customer response time, purchase behavior frequency, emotional state, etc. to form a customer feature vector input recommendation model, which can dig deeper into customer needs. This fusion processing of multi-dimensional data makes outbound call strategies more accurate, provides enterprises with intelligent and customized marketing solutions, and effectively improves sales conversion rates.

[0016] By comprehensively analyzing the customer's response time, purchase tendency and emotional feedback, the present invention can significantly improve the customer's response rate and the conversion effect of outbound calls. The system prioritizes outbound calls to customers with high response rates and provides differentiated communication strategies for customers with different characteristics, thereby improving customer response rates, customer conversion rates and repurchase rates, and enhancing the company's market competitiveness.

[0017] Since the system of the present invention can design personalized outbound call content according to the emotional state and behavior pattern of the customer, the customer experience is improved. The customer obtains recommended information that meets his or her needs during the communication, avoiding negative emotions caused by mismatched communication content, further reducing customer churn and enhancing customer loyalty. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to facilitate understanding by those skilled in the art, the present invention is further described below in conjunction with the accompanying drawings; Figure 1 The figure is a schematic diagram of a call recommendation method for an outbound call system in the present invention.

[0019] Figure 2 The schematic diagram of a call recommendation system for an outbound call system in the present invention. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] Reference Figure 1 - Figure 2 The following examples are obtained: Example

[0022] A call recommendation method for an outbound call system comprises the following steps: The outbound call system obtains customer information and historical interaction records, and pre-processes the data to ensure data integrity and accuracy; after obtaining customer information and historical interaction records, the data is pre-processed to remove invalid data, fill in missing data, standardize the format, and verify the accuracy of the data. This data cleaning process can improve data quality, provide a solid foundation for subsequent analysis and recommendations, reduce noise interference, and make the output of outbound call recommendations more valuable for reference.

[0023] Divide the acquired customer data into several sub-datasets, and classify each sub-dataset based on different customer characteristics; dividing customer data by characteristics can more accurately focus on the personalized needs of customers. Different customers may have differences in response time, purchase frequency, and emotional state. By dividing the data into sub-datasets, the system can classify them based on these characteristics to form customer groups with similar behaviors or characteristics. In this way, personalized outbound call strategies can be designed for each sub-dataset to improve the effectiveness of calls and customer response rates.

[0024] The customer behavior data of each sub-dataset is feature extracted, and the extracted customer feature vector is input into the pre-trained recommendation model to generate a recommendation index based on the customer's behavioral characteristics; the feature extraction step is an important analysis link in outbound call recommendations. The customer's behavioral feature vector can be generated by quantitatively analyzing the customer's response behavior, purchase record, and emotional state. This vector can comprehensively summarize the customer's characteristics in terms of response speed, purchase activity, and emotional fluctuations. After inputting it into the recommendation model, the model will generate a recommendation index based on these characteristics, thereby reflecting the customer's value and priority. The recommendation index provides a reference for the outbound call system in resource allocation and customer prioritization.

[0025] For each customer recommendation index, all customer recommendation results are merged and sorted, and a personalized outbound recommendation plan is finally generated. The recommendation indexes of all customers are aggregated and sorted to form a complete recommendation plan. By sorting the recommendation index, the system can prioritize customers with higher recommendation indexes, that is, those who are more likely to respond and bring value, to avoid wasting resources. In addition, personalized outbound recommendation plans can optimize the efficiency and effectiveness of outbound calls, enabling the outbound call team to achieve better communication results and customer conversion rates in a shorter period of time.

[0026] Extract the basic information of customers from the customer relationship management database, combine it with the historical interaction records of customers extracted from the customer interaction information management database, form a preliminary customer data set, and then perform preprocessing operations on the preliminary customer data set according to the preset preprocessing strategy, store the preprocessed customer data in the comprehensive database and create an index. Retrieve the basic information of customers legally obtained from the customer relationship management database, including the customer's contact number, address, age, gender, tags, etc., which are used to identify and understand the basic background of the customer. Extracting basic customer information can provide basic data for subsequent customer feature analysis, enabling the system to identify customer features and formulate personalized recommendation strategies. Extract the historical interaction records of customers from the customer interaction information management database, such as the date and time of each call, the customer's response time, purchase frequency, feedback content, and emotional expression. Extracting interaction records helps to understand the customer's past behavior and preferences. By analyzing these records, the system can identify the customer's response pattern, emotional state, etc., thereby supporting personalized call recommendations.

[0027] Combine the basic customer information and the customer's historical interaction records to form a preliminary data set containing detailed customer data. The preliminary data set formed by combining basic information and interaction records lays the data foundation for subsequent data processing and feature extraction. With this complete data set, the system can analyze the customer's behavior patterns more comprehensively and lay the foundation for personalized recommendations.

[0028] Preprocess the customer data set according to the preset preprocessing strategy. Specific steps: Preprocess the preliminary data set in the following ways: Data cleaning: Remove invalid data, such as blank records, duplicate information, and unresponsive records. Ensure that each record in the data set provides valid information. Data format standardization: Standardize the formats of all data fields, such as unifying dates into year-month-day format and normalizing the format of phone numbers. Missing value processing: For missing data in key fields (such as customer contact information or preference information), missing data can be filled by supplementing historical data or using estimation methods. Data deduplication: Identify and delete duplicate customer records to ensure the uniqueness of the data and avoid affecting subsequent analysis. Data verification: Verify the accuracy of information, including verifying the validity of the phone number, the correctness of the address, the rationality of the age range, etc. Data preprocessing can improve the integrity, accuracy, and consistency of the data, enabling the system to perform subsequent feature analysis and recommendations based on clean and structured data. Preprocessed data is more reliable, which can reduce noise in subsequent analysis and improve the accuracy of recommendation results.

[0029] The pre-processed customer data is stored in the comprehensive database and indexed. The specific steps are as follows: the pre-processed customer data is stored in the comprehensive database, and a quick search index is established based on the customer's basic information (such as customer number, contact information, address, etc.) to facilitate subsequent inquiries and outbound calls. Storing and indexing data can improve the efficiency of data access, enabling the system to quickly locate specific customer information and records, and improve outbound call efficiency. In addition, the comprehensive database can provide support for data sharing in different modules, achieving more efficient customer information management and outbound call recommendations.

[0030] Dividing the acquired customer data into several sub-datasets, and classifying each sub-dataset based on different customer characteristics means: according to the recommendation requirements of the outbound call system, selecting the core feature dimensions of the customer data for division, the core feature dimensions include customer response time, purchase behavior frequency and call emotional state; setting classification standards for each feature dimension, specifically including: dividing the customer's average response time into fast response customers, medium response customers and slow response customers according to a preset threshold; dividing customers into high-frequency purchase customers, medium-frequency purchase customers and low-frequency purchase customers according to their purchase records; dividing customers into high-emotion response customers, neutral-emotion response customers and low-emotion response customers according to emotional scores; based on the above feature classification standards, dividing the customer data into several sub-datasets, each sub-dataset corresponds to a specific single feature category or combined feature category, and assigning a unique identifier to each divided sub-dataset for subsequent retrieval.

[0031] Dividing customer data into different sub-datasets based on core features (such as response time, purchase frequency, and emotional state) can effectively distinguish customer behaviors and preferences. For example, customers who respond quickly are more likely to answer the phone, while customers who buy frequently may be more interested in new products. After classifying customers by characteristics, outbound call strategies can be more accurately formulated based on the behaviors of each group, thereby allocating outbound call resources in a more targeted manner.

[0032] Different customer groups have different preferences in communication methods and content. By dividing the sub-datasets, the outbound call content and methods can be tailored for each characteristic group. For example, for customers with high emotional responses, a gentler communication method can be used, while for customers with slow responses, a more positive method can be tried or follow-up notifications can be sent. Personalized outbound calls can better meet customer expectations and thus improve customer satisfaction.

[0033] After dividing the sub-datasets, the outbound call system can allocate outbound call resources according to the priority of each sub-dataset. For example, customers with high frequency of purchase or quick response may be called first, while customers with low frequency of purchase or low emotional response can be called when resources are free. In this way, resources can be concentrated on customers who are most likely to generate conversions, avoiding resource waste.

[0034] After the sub-datasets are divided, each customer group has different priorities and outbound call recommendation indexes. The outbound call system can prioritize customer data sets with high recommendation indexes and gradually lower the priority for calls. In this way, the system can manage the outbound call sequence in an orderly manner, avoid the chaos of disordered calls, and improve the success rate of outbound calls.

[0035] Dividing sub-datasets can also facilitate data management and analysis. For example, for a specific type of customer group, the outbound call system can track and analyze their responses and behavior trends at any time, and adjust the outbound call strategy in real time based on these analysis results to optimize the strategy. This clear hierarchical structure provides higher efficiency and accuracy for data analysis.

[0036] Dividing the sub-datasets can provide clearer feature inputs for the recommendation model. The customer feature vectors of different sub-datasets can be input into the outbound call recommendation model, and the recommendation model can generate an outbound call plan that is more suitable for the group. This recommendation method can better match customer needs and improve customer response rate.

[0037] Outbound call systems can design differentiated communication strategies based on different sub-data sets. For example, for customers with high emotional responses, senior customer service staff can be given priority for communication, while for customers with low frequency of purchases, promotions or discounts can be used to attract their attention. These personalized strategies increase the attractiveness of outbound calls and the possibility of customer conversion.

[0038] Furthermore, the following optimizations can be performed: During outbound calls, the system can evaluate the communication effect in real time based on the customer response data of the sub-datasets. For example, if the feedback from customers with high emotional response is good, the call intensity to this group can be increased; if the response from medium-frequency purchasing customers is low, the preferential strategy or communication method can be adjusted in the next outbound call. Dividing the sub-datasets also makes it easier to record subsequent customer feedback and responses by group. The system can summarize the preferences and needs of different customer groups and feed these data back to the recommendation model to continuously optimize the calculation of the recommendation index and the formulation of outbound call strategies.

[0039] Feature extraction of customer behavior data in each sub-dataset refers to extracting customer response information, customer purchase information and customer call emotion information respectively, and then performing feature analysis operations to generate customer response index, customer purchase index and customer emotion index respectively, and finally aggregating the customer response index, customer purchase index and customer emotion index together into a customer feature vector.

[0040] The system extracts the customer's response information from the customer's historical interaction records, calculates the customer's average response speed, consistency and regularity, and uses this information to generate a customer response index. The customer response index reflects the customer's sensitivity and response tendency to outbound calls. Customers with fast and consistent response speed may be more active in outbound calls, while customers with slower responses require different outbound call strategies. The system extracts the customer's purchase information from the customer's purchase record, including the frequency of purchases, the amount of each purchase, consumption fluctuations, etc. The customer's purchase activity, consumption level, consumption fluctuations and purchase trends are calculated to generate a customer purchase index. The customer purchase index represents the customer's consumption behavior and purchase intention. Customers with a high purchase index usually have a strong willingness to buy and are suitable for priority promotion or recommendation, while customers with a low purchase index may need more guidance and incentives. Emotional state data is extracted from the customer's historical call records, including the amplitude of emotional changes, emotional stability and recent emotional trends. The extremeness, volatility and stability of customer emotions are calculated, and the customer emotion index is generated based on this information. The customer emotion index can reflect the customer's acceptance of communication. Customers with more volatile emotions may be sensitive to the content of communication, while customers with stable emotions may be more friendly to communication, thus affecting the communication method of the recommendation strategy.

[0041] The customer response index, customer purchase index, and customer sentiment index are aggregated to generate a complete customer feature vector. The various indexes are combined to form a comprehensive feature vector that represents the overall characteristics of the customer. The customer feature vector can fully reflect the customer's behavior patterns and preferences, and is the core input for the recommendation model to generate the recommendation index. Based on these comprehensive features, the system can more accurately formulate personalized outbound call plans and improve customer response rates and conversion rates.

[0042] The logic of obtaining the customer response index is: extract the time data of each response from the customer response information, and define the customer response time series as The timestamp represents the i-th response, and n is the total number of responses; this sequence is used to record the customer's response time in each interaction, forming the basis for calculating response speed and consistency. By recording each response time point, the customer's response pattern, speed, regularity and other characteristics can be analyzed later as raw data for index calculation.

[0043] According to the customer's response time series, calculate the time interval between two adjacent responses to obtain the response time interval sequence Represents the time difference between the i-th and i+1-th responses; the time interval sequence is used to calculate the average speed and consistency of responses. This information can characterize the customer's response habits and frequency in multiple interactions, which helps to judge the customer's activity.

[0044] Compute response time interval sequence The mean , reflecting the average response speed of customers; the average response speed can be used to identify the response tendency of customers. A shorter average response time means that the customer responds quickly and is suitable for priority outbound calls, while a longer response time indicates that the customer responds slowly.

[0045] Calculate the variance of the response time interval series , which measures the consistency of customer responses; smaller variances indicate more stable response times, while larger variances indicate unstable responses. Consistency reflects the customer's response pattern during the interaction. Stable response patterns help predict the customer's future response behavior, while inconsistent patterns indicate that the customer's responses are more volatile.

[0046] Calculate the autocorrelation function of the response time interval series at the lag k time , used to evaluate the periodicity and regularity of the response interval, the formula is: ; The autocorrelation function is used to identify potential periodicity in customer responses. For example, customers may respond after a certain time interval, and this pattern can help the system predict customer response patterns and make outbound calls more effectively.

[0047] The customer response index is calculated based on the mean, variance, and autocorrelation function of the response time interval and is defined as: ; K represents the maximum time point of the autocorrelation function, which is used to balance the short-term and long-term response laws. are preset weighting coefficients, which are used to adjust the weights of mean, variance and autocorrelation function in the customer response index. is the maximum value of the variance, which is used to standardize the variance term, and R is the customer response index. The customer response index R is an important indicator to measure the customer's response tendency. It takes into account the speed, consistency and regularity of the response to help the outbound call system identify the priority of customers. Customers with a higher response index are more likely to respond first and are suitable for priority outbound calls, while customers with a lower response index may require different outbound call strategies.

[0048] The logic for obtaining the customer purchase index is: Extract the time and amount data of each purchase from the customer purchase information and define the customer purchase record sequence as The timestamp of the i-th purchase is represented by the corresponding purchase amount, and m is the total number of purchase records; this sequence records the customer's purchase behavior and provides raw data for subsequent analysis of the customer's consumption frequency, consumption amount fluctuations, etc. By fully recording the customer's purchase time and amount data, it is possible to analyze the customer's consumption behavior pattern and lay the foundation for the calculation of the purchase index.

[0049] Calculate the average frequency of purchase intervals based on the customer's purchase record sequence , reflects the customer's purchasing activity and calculates the average purchase amount , reflecting the average consumption level of customers and calculating the variance of purchase amount , which is used to measure the volatility of customer spending; frequency reflects the level of activity of customer purchasing behavior. Customers with high frequency usually have a stronger willingness to consume, so it is suitable to give priority to recommending new products or services when making outbound calls. The average consumption level can be used to evaluate the customer's spending ability and willingness. A higher mean means that the customer invests more in consumption, and the outbound call system can provide high-value products or services to this customer. Variance provides information on the stability of customer consumption behavior. Customers with large fluctuations may have high consumption behavior in specific periods of time, and these high-consumption opportunities can be seized through outbound call recommendations, while customers with smaller consumption fluctuations tend to have stable consumption habits.

[0050] The customer's purchase trend coefficient is calculated using the time-weighted average method to reflect the trend of the customer's consumption habits changing over time. The calculation formula is: is the purchase trend coefficient, which indicates the influence of the customer's recent consumption habits on the purchase index; the closer the purchase behavior is to the current time, the greater the influence on the coefficient. The purchase trend coefficient can help identify the direction of change in the customer's recent consumption behavior. If the customer is relatively high, it means that the customer has a high consumption activity recently, and this trend can be used to follow up in time; if If the consumption rate gradually decreases, it may be necessary to adjust the recommendation strategy to attract customers to continue consuming.

[0051] Based on the purchase frequency, the mean, variance and purchase trend coefficient of the purchase amount, the customer purchase index is calculated as follows: is the maximum value of the purchase amount variance, which is used to standardize the variance. They are the influence coefficients of purchase frequency, average consumption, consumption stability and purchase trend, and D is the customer purchase index. The customer purchase index D is an important indicator to measure customer purchase behavior. Through the comprehensive calculation of activity, consumption level, consumption stability and trend, it helps the outbound call system identify customer groups with higher priority. Customers with higher indexes are suitable for giving priority to recommending new products or services during outbound calls to improve sales conversion rate. The logic for obtaining the customer sentiment index is: Extract the emotional state of each call from the customer's call emotional information and define the emotional state sequence as Indicates the emotional state of the ith call, with a value range of [0,1], indicating the intensity of emotions from negative to positive, and v is the total number of emotion records; the emotional state sequence is the basis for calculating customer emotional characteristics. By recording the emotional changes of customers in each call, we can analyze the emotional patterns and emotional fluctuations of customers during the communication process, thereby helping the outbound call system understand the emotional tendencies of customers.

[0052] The maximum and minimum emotional state values ​​reflect the extreme bias of customer emotions. The calculation formula is: It is an extreme bias value, which is used to quantify the extreme volatility of the customer's emotional state. This value quantifies the extreme volatility of the customer's emotions. The extreme volatility of emotions can indicate the volatility and extremeness of the customer's emotions. If the customer's emotions fluctuate greatly, a more cautious communication strategy may be required in outbound calls, while customers with stable emotions are more likely to accept outbound call content.

[0053] The exponential function is used to perform nonlinear gain processing on the emotion change to enhance the impact of customer emotion fluctuations on the emotion index. The formula is: The emotional gain value is the average of the difference between two consecutive emotional states, which reflects the volatility of customer emotions. The higher the gain value, the greater the change in customer emotions. The emotional gain value can reflect the emotional ups and downs of customers. If the customer's emotions change drastically, the outbound call system may need to adjust the communication content to adapt to the changes in customer emotions to avoid triggering negative reactions.

[0054] The customer sentiment index calculation formula is: ; ; is the preset attenuation factor, are all preset emotional infection coefficients, and F is the customer emotional index. When defining the time decay weight factor wi, the weight factor is used to assign different weights to emotional states at different times, so that the most recent emotional state has a greater impact on the emotional index. Through the method of time decay, it is ensured that the customer's recent emotional state has a greater influence when calculating the emotional index. This can help the outbound call system understand the customer's current emotional state and adjust the communication strategy in a timely manner. The customer emotional index F quantifies the customer's overall emotional state, including emotional extremes, volatility, and recent emotional impacts. The higher the emotional index, the more positive and stable the customer's emotions are, which is suitable for outbound communication. Customers with a lower index may have negative emotions and need to adjust the communication method to avoid negative reactions.

[0055] The recommendation model is obtained through convolutional neural network training. When used, the customer response index, customer purchase index and customer emotion index are summarized as a customer feature vector and input into the recommendation model. The output of the recommendation model is the recommendation index. The response index, purchase index and emotion index of each customer are taken as independent feature values, and these feature values ​​are combined into a complete feature vector. The customer's response index represents the customer's response speed and consistency to outbound calls, the purchase index reflects the customer's consumption tendency and activity, and the emotion index reflects the customer's emotional tendency and fluctuation during the communication process. Integrating these three indexes into a feature vector can comprehensively characterize the characteristics of each customer and provide complete input data for the model.

[0056] In the recommendation model, the customer feature vector is used as input to the trained convolutional neural network. The convolutional neural network performs deep feature extraction on the customer feature vector and identifies the potential patterns of customer behavior and preferences. By extracting features and identifying patterns on customer feature vectors, the convolutional neural network model can capture the mutual influence between different features. For example, active customers may respond more positively to outbound calls, and customers with a higher sentiment index may be more receptive to recommendations.

[0057] After the convolutional neural network processes the customer feature vector, it generates a recommendation index. The recommendation index represents the system's prediction of the likelihood and value of the customer responding to the outbound call. The recommendation index is one of the core outputs of the outbound call system, quantifying the customer's acceptance and conversion potential of outbound recommendations. Customers with a higher index have a higher outbound call priority and are suitable for providing higher-value recommendations during outbound calls, while customers with a lower index may require different communication strategies or be ranked in the priority of subsequent calls.

[0058] Example 2 A call recommendation system for an outbound call system, comprising: The data acquisition and processing module obtains customer information and historical interaction records through the outbound call system, pre-processes the data to ensure the integrity and accuracy of the data, and stores the pre-processed customer data in the comprehensive database; The customer data segmentation module divides the customer data obtained from the comprehensive database into several sub-data sets, and each sub-data set is classified based on different customer characteristics; The feature extraction module extracts features from the customer behavior data of each sub-dataset, uses the extracted customer feature vector to input into the pre-trained recommendation model, and generates a recommendation index based on the customer's behavior characteristics; The outbound call recommendation plan generation module combines and sorts the recommendation results of all customers for each customer recommendation index, and finally generates a personalized outbound call recommendation plan.

[0059] The sorting is based on the customer recommendation index. The final personalized outbound recommendation plan is based on the customer classification results. An example is as follows: Example 1: Prioritize response to customer outbound referral plan: Customer sub-dataset classification: fast response + high frequency purchase + high emotional response; fast response + medium frequency purchase + high emotional response; fast response + high frequency purchase + neutral emotional response Sorting priority: Since these customers belong to the "Quick Response" category, their recommendation index is prioritized, and the system will further sort them based on the customer's purchase frequency and emotional state.

[0060] Personalized content: Quick response + high frequency purchase + high emotional response customers: Provide new products or special offers to increase the chance of interaction. Quick response + medium frequency purchase + high emotional response customers: Provide cost-effective recommendations to attract customers to consume. Quick response + high frequency purchase + neutral emotional response customers: Provide some discount information to stimulate customers' purchasing interest.

[0061] Sample referral plan: Customer A: Quick response + high frequency purchase + high emotional response, recommendation index is 98 (highest priority), outbound call content: invitation to participate in new product launch conference and provide exclusive discount offers.

[0062] Customer B: Quick response + medium-frequency purchase + high emotional response, recommendation index is 90, outbound call content: recommend a popular product with a small discount.

[0063] Customer C: Quick response + high frequency purchase + neutral emotional response, recommendation index is 85, outbound call content: introduce recent promotions and provide free trial opportunities. By giving priority to calling these quick-responding customers, the outbound call system can obtain a higher feedback rate in a shorter period of time, and these customers have a positive attitude towards communication, so the success rate of outbound calls is higher.

[0064] Example 2: Outbound call recommendation plan for slow-responding customers: Customer sub-dataset classification: slow response + low-frequency purchase + high emotional response; slow response + medium-frequency purchase + neutral emotional response; slow response + low-frequency purchase + low emotional response Since these customers have a slower response time and lower ranking priority, the outbound call system will call these customers after the high-priority and medium-priority customers have been processed.

[0065] Personalized content: Slow response + low frequency purchase + high emotional response customers: You can use more attractive content, such as limited-time discounts or gifts, to attract customers' interest. Slow response + medium frequency purchase + neutral emotional response customers: Provide a practical product or service to attract their attention. Slow response + low frequency purchase + low emotional response customers: Choose to send messages or low-frequency outbound calls to avoid frequent interruptions.

[0066] Sample referral plan: Customer D: slow response + low frequency purchase + high emotional response, recommendation index is 65 (low priority), outbound call content: provide limited-time discounts and attach small gifts to stimulate customer interest in participation.

[0067] Customer E: Slow response + medium-frequency purchase + neutral emotional response, recommendation index is 60, outbound call content: recommend a cost-effective product to meet its basic needs.

[0068] Customer F: Slow response + low frequency of purchase + low emotional response, recommendation index is 50, outbound call content: send notification information, introduce current activity information, and avoid frequent phone calls. For slow response customers, the outbound call recommendation plan avoids disturbing customers through personalized content design and low frequency outbound call strategy, and attracts customers' attention through moderate discounts to increase the possibility of response.

[0069] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0070] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0071] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0072] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0073] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A call recommendation method for an outbound call system, characterized in that: The following steps are involved: Obtain customer information and historical interaction records through the outbound call system, and pre-process the data to ensure its integrity and accuracy; Divide the acquired customer data into several sub-datasets, and classify each sub-dataset based on different customer characteristics; Extract features from the customer behavior data of each sub-dataset, input the extracted customer feature vector into the pre-trained recommendation model, and generate a recommendation index based on the customer's behavior characteristics; For each customer recommendation index, all customer recommendation results are merged and sorted, and ultimately a personalized outbound recommendation plan is generated.

2. A call recommendation method for an outbound call system according to claim 1, characterized in that: The basic information of customers is extracted from the customer relationship management database, combined with the customer historical interaction records extracted from the customer interaction information management database to form a preliminary customer data set, and then the preliminary customer data set is preprocessed according to the preset preprocessing strategy, and the preprocessed customer data is stored in the comprehensive database and indexed.

3. A call recommendation method for an outbound call system according to claim 2, characterized in that: Dividing the acquired customer data into several sub-datasets, each of which is classified based on different customer characteristics, means: According to the recommendation requirements of the outbound call system, the core feature dimensions of customer data are selected for segmentation. The core feature dimensions include customer response time, purchase behavior frequency, and call emotional state. Set classification standards for each feature dimension, including: The average response time of customers is divided into fast response customers, medium response customers and slow response customers according to a preset threshold; According to the customer's purchase records, customers are divided into high-frequency purchase customers, medium-frequency purchase customers and low-frequency purchase customers; Divide customers into high emotional response customers, neutral emotional response customers, and low emotional response customers based on their emotional scores; Based on the above feature classification standards, the customer data is divided into several sub-datasets, each of which corresponds to a specific single feature category or combined feature category, and a unique identifier is assigned to each divided sub-dataset to facilitate subsequent retrieval.

4. A call recommendation method for an outbound call system according to claim 3, characterized in that: Feature extraction of customer behavior data in each sub-dataset refers to extracting customer response information, customer purchase information and customer call emotion information respectively, and then performing feature analysis operations to generate customer response index, customer purchase index and customer emotion index respectively, and finally aggregating the customer response index, customer purchase index and customer emotion index together into a customer feature vector.

5. A call recommendation method for an outbound call system according to claim 4, characterized in that: The logic of obtaining the customer response index is: extract the time data of each response from the customer response information, and define the customer response time series as , Indicates the timestamp of the i-th response, and n is the total number of responses; According to the customer's response time series, calculate the time interval between two adjacent responses to obtain the response time interval sequence , represents the time difference between the i-th and i+1-th responses; Compute response time interval sequence The mean , reflects the average response speed of customers; calculates the variance of the response time interval series , measure the consistency of customer responses; calculate the autocorrelation function of the response time interval series at the lag k moment , used to evaluate the periodicity and regularity of the response interval, the formula is: ; The customer response index is calculated based on the mean, variance and autocorrelation function of the response time interval, which is defined as: ; K represents the maximum time point of the autocorrelation function, which is used to balance the short-term and long-term response laws are preset weighting coefficients, which are used to adjust the weights of mean, variance and autocorrelation function in the customer response index. is the maximum value of the variance, which is used to standardize the variance term, and R is the customer response index.

6. A call recommendation method for an outbound call system according to claim 5, characterized in that: The logic for obtaining the customer purchase index is: extract the time and amount data of each purchase from the customer purchase information, and define the customer purchase record sequence as , represents the timestamp of the i-th purchase, represents the corresponding purchase amount, m is the total number of purchase records; based on the customer's purchase record sequence, calculate the average frequency of the purchase time interval , reflects the customer's purchasing activity and calculates the average purchase amount , reflects the average consumption level of customers and calculates the variance of purchase amount , used to measure the volatility of customer spending; the customer's purchase trend coefficient is calculated using the time-weighted average method to reflect the trend of customer spending habits over time. The calculation formula is: ; is the purchase trend coefficient, which indicates the influence of the customer’s recent consumption habits on the purchase index; Based on the purchase frequency, the mean, variance and purchase trend coefficient of the purchase amount, the customer purchase index is calculated as follows: is the maximum value of the purchase amount variance, which is used to standardize the variance. They are the influence coefficients of purchase frequency, average consumption, consumption stability and purchase trend, and D is the customer purchase index.

7. A call recommendation method for an outbound call system according to claim 6, characterized in that: The logic for obtaining the customer sentiment index is: Extract the emotional state of each call from the customer's call emotional information and define the emotional state sequence as It represents the emotional state of the i-th call, with a value range of [0,1], indicating the emotional intensity from negative to positive, and v is the total number of emotional records; the maximum and minimum emotional state values ​​reflect the extreme bias of customer emotions, and the calculation formula is: It is an extreme bias value, which is used to quantify the extreme volatility of the customer's emotional state. The exponential function is used to perform nonlinear gain processing on the emotional change to enhance the impact of customer emotional fluctuations on the emotional index. The formula is: is the mood gain value; The customer sentiment index calculation formula is: ; ; is the preset attenuation factor, are all preset emotion infection coefficients, and F is the customer emotion index.

8. A call recommendation method for an outbound call system according to claim 7, characterized in that: The recommendation model is obtained through convolutional neural network training. When used, the customer response index, customer purchase index and customer sentiment index are summarized as a customer feature vector and input into the recommendation model. The output result of the recommendation model is the recommendation index.

9. A call recommendation system for an outbound call system, implemented based on a call recommendation method for an outbound call system according to any one of claims 1 to 8, characterized in that: include: The data acquisition and processing module obtains customer information and historical interaction records through the outbound call system, pre-processes the data to ensure the integrity and accuracy of the data, and stores the pre-processed customer data in the comprehensive database; The customer data segmentation module divides the customer data obtained from the comprehensive database into several sub-data sets, and each sub-data set is classified based on different customer characteristics; The feature extraction module extracts features from the customer behavior data of each sub-dataset, uses the extracted customer feature vector to input into the pre-trained recommendation model, and generates a recommendation index based on the customer's behavior characteristics; The outbound call recommendation plan generation module combines and sorts the recommendation results of all customers for each customer recommendation index, and finally generates a personalized outbound call recommendation plan.

Images