Call behavior time sequence attenuation-based outbound call contact intention prediction method and system

Abstract

The invention discloses an outbound call contact intention prediction method and system based on call behavior time sequence attenuation, and the method comprises the steps: obtaining call behaviors of clients through an outbound call log, and generating a contact event sequence for the call behaviors of each client; based on a preset feature extraction algorithm, extracting behavior feature parameters corresponding to each contact event sequence, and constructing a behavior feature table; determining willingness attenuation parameters according to the behavior feature table and the contact event sequence, and constructing a mixed willingness attenuation model; and according to the contact event sequence of the target customer and the mixed willingness attenuation model, calculating an outbound contact willingness score of the target customer. Therefore, accurate outbound contact intention scoring based on the individual intention attenuation trend and the group intention attenuation trend can be realized, the rationality and the reach rate of an outbound contact strategy are improved, and the influence of neglecting of customer contact intentions on the reduction of the working efficiency of disposition of unfavorable assets is reduced.

Description

Technical Field

[0001] This invention relates to the field of internet finance technology, and in particular to a method and system for predicting outbound call contact intentions based on the temporal decay of call behavior. Background Technology

[0002] In internet finance, creditors remind debtors that the repayment of non-performing assets relies on high-frequency outbound calling strategies. However, traditional outbound calling strategies lack the ability to dynamically identify and respond to debtors' willingness to be contacted, and often require the determination of outbound calling strategies based on debtor profiles.

[0003] In order to achieve debtor repayment follow-up processing, existing technologies have proposed repayment follow-up strategies based on debtor profile assessment. For example, patent publication number CN120833213A, entitled "Collection Strategy Generation and Execution System and Method Based on Debtor Profile," discloses the determination of collection allocation processing strategies under each user profile, optimization of the collection processing order and collection processing allocation scheme under each user profile, thereby improving the recognition and processing efficiency when there are deviations in the emotional recognition under each user profile.

[0004] However, the existing technologies still have the following shortcomings: debtor repayment follow-up is highly dependent on the outbound call contact status, and the analysis of debtor willingness is based solely on numerical indicators of call data, without considering the dynamic decline in the debtor's psychological tolerance threshold during continuous contact. As the number of contact interruptions increases and the interval between repeat calls shortens, the debtor's willingness to contact decreases rapidly and non-linearly. This fails to provide an interpretable, calculable, and deployable transparent logical expression for predicting debtor contact willingness. Especially in the compliance-critical context of internet finance, the lack of theoretical rigor and engineering practicality in willingness prediction will affect the efficiency and compliance of non-performing asset disposal. Therefore, the existing technologies have shortcomings that urgently need to be addressed. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a method and system for predicting outbound contact intention based on the temporal decay of call behavior. This method can achieve accurate outbound contact intention scoring based on individual and group intention decay trends, improve the rationality and reach of outbound contact strategies, and reduce the impact of neglecting customer contact intention on the efficiency of non-performing asset disposal.

[0006] To address the aforementioned technical problems, the first aspect of this invention discloses a method for predicting outbound call contact intentions based on call behavior temporal attenuation, the method comprising: Customer call behavior is obtained through outbound call logs, and a contact event sequence is generated for each customer's call behavior. Based on a preset feature extraction algorithm, behavioral feature parameters corresponding to each contact event sequence are extracted to construct a behavioral feature table. Based on the behavioral feature table and the contact event sequence, the intention decay parameters are determined, and a hybrid intention decay model is constructed. Based on the target customer's contact event sequence and the hybrid intention decay model, the outbound contact intention score of the target customer is calculated.

[0007] As an optional implementation, in the first aspect of the invention, before obtaining the customer's call behavior through outbound call logs, the following steps are included: Outbound call logs are collected and cleaned and standardized. The outbound call logs include customer identifiers, time sequence identifiers, and fields that record the call behavior. The data cleaning process removes valid outbound calls from the outbound call logs while retaining invalid outbound call data. The standardization process unifies the expression of the call behavior by the fields.

[0008] As an optional implementation, in the first aspect of the present invention, the call behavior includes outbound call interruption, call retry delay, answer interval response, cumulative outbound calls, outbound call rejection, and outbound call complaint; The length of the contact event sequence is the number of outbound calls made by the customer for the corresponding call behavior.

[0009] As an optional implementation, in the first aspect of the present invention, the behavioral characteristic parameters include interruption parameters, repeat call interval parameters, outbound call frequency parameters, historical rejection parameters, and complaint parameters. The interruption parameter is the cumulative number of call actions corresponding to the customer's call interruption within and outside the contact event sequence; The repeat call interval parameter is the time interval between the outbound call at the first moment and the outbound call at the second moment within the contact event sequence; The outbound call frequency parameter is the frequency of outbound call behavior within the contact cycle of the contact event sequence; The historical rejection parameter is the rejection rate of a customer within the contact period of a contact event sequence. The rejection rate is the ratio between the number of rejection behaviors and the total number of outbound calls. The rejection behaviors specifically include outbound call rejection and / or outbound call non-answer. The complaint parameter refers to the status value corresponding to whether the customer has complained to the blacklist within the contact event sequence.

[0010] As an optional implementation, in the first aspect of the present invention, the willingness decay parameter includes individual initial willingness, interruption sensitivity coefficient, delay recovery rate, and complaint interception index; The individual initial intention is used to characterize the differences in individual intentions of any customer in the initial contact in a sequence of contact events; The interruption sensitivity coefficient is used to characterize the intensity of the impact of each increase in outbound call interruption on the decline in contact willingness; The delayed recovery rate is used to characterize the ability of any customer to resume contact during the outbound call interval; The complaint interception metric is used to characterize any customer's willingness to be contacted regarding a complaint or non-complaint. The individual's initial intention and the interruption sensitivity coefficient are associated with any customer's contact event sequence, which constitutes the driving attenuation term of the outbound call interruption behavior pattern. The delayed recovery rate is associated with the sequence of contact events for any customer, forming a delayed recovery factor term in the response interval response behavior pattern. Based on the driving attenuation term and the delay recovery factor term, a nonlinear model is performed to explicitly model the nonlinear relationship between the customer's outbound call interruption behavior pattern and the response interval response behavior pattern in the contact event sequence.

[0011] As an optional implementation, in the first aspect of the invention, constructing the hybrid intention decay model includes: A loss-of-contact risk model is constructed based on survival analysis. The behavioral feature table and the contact event sequence of any customer are obtained and input into the loss-of-contact risk model to obtain the loss-of-contact risk value of the corresponding customer to distinguish risky customers. The loss-of-contact risk model is trained by the behavioral feature table and the contact event sequence. The behavioral feature table and the contact event sequence of any customer are obtained as input. Based on Bayesian exponential decay fitting, the individual initial intention of the intention decay parameter and the repeat call threshold corresponding to the individual initial intention are obtained. The behavioral feature table and the aggregated contact event sequence are used as inputs, and the interruption sensitivity coefficient and delay recovery rate of the intention decay parameter are obtained by sliding window estimation fitting. The system obtains a sequence of contact events for any customer as input, and dynamically captures the decay of contact intention in the sequence of contact events based on a neural network prediction model to predict the customer's willingness to call back at the last moment. The neural network prediction model is trained using the behavioral feature table and the sequence of contact events. The system obtains a contact event sequence of any customer as input, and uses the LightGBM model to perform auxiliary estimation of the repeat contact intention score on the behavioral feature parameters in the contact event sequence, so as to help correct the repeat contact intention score. The LightGBM model is trained by the behavioral feature table, the contact event sequence and the labeled repeat contact intention score.

[0012] As an optional implementation, in the first aspect of the present invention, the construction of the hybrid intention decay model further includes: The intention decay function is modeled based on the lost contact risk model and the Bayesian exponential decay fitting. The intention decay function is then probabilistically modeled and nondeterministically quantified based on the Bayesian exponential decay fitting. Finally, the customer's outbound contact intention score is predicted based on the intention decay function using the LightGBM model and a neural network prediction model, resulting in the hybrid intention decay model.

[0013] As an optional implementation, in the first aspect of the present invention, the calculation expression for the target customer's outbound contact willingness score is: ; ; ; in, This indicates that the hybrid intention decay model calculates the target customer's initial contact intention score. This indicates that the auxiliary correction model predicts and estimates the target customer's second contact intention score. This indicates the target customer's willingness to be contacted via outbound calls. This represents the initial individual intention of the target customer. Represents the natural constant. This represents the interruption sensitivity coefficient. This indicates the interruption parameters for the target customer. Indicates the latency recovery rate of the target customer. This represents the repeat call interval parameter for the target customer. This indicates the metrics for intercepting complaints from target customers. This represents the intention estimation function. This parameter represents the outbound call frequency within a period d of the target customer contact event sequence. This indicates the historical rejection parameters of the target customer. The score weights represent the predictions of declining willingness. Let i represent the sequence number of the target customer i, and t represent the t-th outbound call in the contact event sequence of the target customer.

[0014] A second aspect of this invention discloses an outbound call contact intention prediction system based on call behavior temporal attenuation, the system comprising: The acquisition module is used to acquire customers' call behavior through outbound call logs and generate a contact event sequence for each customer's call behavior; The extraction module is used to extract behavioral feature parameters corresponding to each contact event sequence based on a preset feature extraction algorithm, and to construct a behavioral feature table. The determination module is used to determine the intention decay parameters based on the behavioral feature table and the contact event sequence, and to construct a hybrid intention decay model; The calculation module is used to calculate the outbound contact willingness score of the target customer based on the contact event sequence of the target customer and the hybrid willingness decay model.

[0015] As an optional implementation, in a second aspect of the invention, before obtaining the customer's call behavior through outbound call logs, the following steps are included: Outbound call logs are collected and cleaned and standardized. The outbound call logs include customer identifiers, time sequence identifiers, and fields that record the call behavior. The data cleaning process removes valid outbound calls from the outbound call logs while retaining invalid outbound call data. The standardization process unifies the expression of the call behavior by the fields.

[0016] As an optional implementation, in a second aspect of the present invention, the call behavior includes outbound call interruption, call retry delay, answer interval response, cumulative outbound calls, outbound call rejection, and outbound call complaint; The length of the contact event sequence is the number of outbound calls made by the customer for the corresponding call behavior.

[0017] As an optional implementation, in the second aspect of the present invention, the behavioral characteristic parameters include interruption parameters, repeat call interval parameters, outbound call frequency parameters, historical rejection parameters, and complaint parameters; The interruption parameter is the cumulative number of call actions corresponding to the customer's call interruption within and outside the contact event sequence; The repeat call interval parameter is the time interval between the outbound call at the first moment and the outbound call at the second moment within the contact event sequence; The outbound call frequency parameter is the frequency of outbound call behavior within the contact cycle of the contact event sequence; The historical rejection parameter is the rejection rate of a customer within the contact period of a contact event sequence. The rejection rate is the ratio between the number of rejection behaviors and the total number of outbound calls. The rejection behaviors specifically include outbound call rejection and / or outbound call non-answer. The complaint parameter refers to the status value corresponding to whether the customer has complained to the blacklist within the contact event sequence.

[0018] As an optional implementation, in the second aspect of the present invention, the willingness decay parameter includes individual initial willingness, interruption sensitivity coefficient, delay recovery rate, and complaint interception index; The individual initial intention is used to characterize the differences in individual intentions of any customer in the initial contact in a sequence of contact events; The interruption sensitivity coefficient is used to characterize the intensity of the impact of each increase in outbound call interruption on the decline in contact willingness; The delayed recovery rate is used to characterize the ability of any customer to resume contact during the outbound call interval; The complaint interception metric is used to characterize any customer's willingness to be contacted regarding a complaint or non-complaint. The individual's initial intention and the interruption sensitivity coefficient are associated with any customer's contact event sequence, which constitutes the driving attenuation term of the outbound call interruption behavior pattern. The delayed recovery rate is associated with the sequence of contact events for any customer, forming a delayed recovery factor term in the response interval response behavior pattern. Based on the driving attenuation term and the delay recovery factor term, a nonlinear model is performed to explicitly model the nonlinear relationship between the customer's outbound call interruption behavior pattern and the response interval response behavior pattern in the contact event sequence.

[0019] As an optional implementation, in a second aspect of the invention, constructing the hybrid intention decay model includes: A loss-of-contact risk model is constructed based on survival analysis. The behavioral feature table and the contact event sequence of any customer are obtained and input into the loss-of-contact risk model to obtain the loss-of-contact risk value of the corresponding customer to distinguish risky customers. The loss-of-contact risk model is trained by the behavioral feature table and the contact event sequence. The behavioral feature table and the contact event sequence of any customer are obtained as input. Based on Bayesian exponential decay fitting, the individual initial intention of the intention decay parameter and the repeat call threshold corresponding to the individual initial intention are obtained. The behavioral feature table and the aggregated contact event sequence are used as inputs, and the interruption sensitivity coefficient and delay recovery rate of the intention decay parameter are obtained by sliding window estimation fitting. The system obtains a sequence of contact events for any customer as input, and dynamically captures the decay of contact intention in the sequence of contact events based on a neural network prediction model to predict the customer's willingness to call back at the last moment. The neural network prediction model is trained using the behavioral feature table and the sequence of contact events. The system obtains a contact event sequence of any customer as input, and uses the LightGBM model to perform auxiliary estimation of the repeat contact intention score on the behavioral feature parameters in the contact event sequence, so as to help correct the repeat contact intention score. The LightGBM model is trained by the behavioral feature table, the contact event sequence and the labeled repeat contact intention score.

[0020] As an optional implementation, in the second aspect of the invention, the construction of the hybrid intention decay model further includes: The intention decay function is modeled based on the lost contact risk model and the Bayesian exponential decay fitting. The intention decay function is then probabilistically modeled and nondeterministically quantified based on the Bayesian exponential decay fitting. Finally, the customer's outbound contact intention score is predicted based on the intention decay function using the LightGBM model and a neural network prediction model, resulting in the hybrid intention decay model.

[0021] As an optional implementation, in the second aspect of the invention, the calculation expression for the target customer's outbound contact willingness score is: ; ; ; in, This indicates that the hybrid intention decay model calculates the target customer's initial contact intention score. This indicates that the auxiliary correction model predicts and estimates the target customer's second contact intention score. This indicates the target customer's willingness to be contacted via outbound calls. This represents the initial individual intention of the target customer. Represents the natural constant. This represents the interruption sensitivity coefficient. This indicates the interruption parameters for the target customer. Indicates the latency recovery rate of the target customer. This represents the repeat call interval parameter for the target customer. This indicates the metrics for intercepting complaints from target customers. This represents the intention estimation function. This parameter represents the outbound call frequency within a period d of the target customer contact event sequence. This indicates the historical rejection parameters of the target customer. The score weights represent the predictions of declining willingness. Let i represent the sequence number of the target customer i, and t represent the t-th outbound call in the contact event sequence of the target customer.

[0022] A third aspect of the present invention discloses another outbound call contact intention prediction system based on call behavior temporal decay, the system comprising: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute some or all of the steps in the outbound contact intention prediction method based on call behavior temporal decay disclosed in the first aspect of the present invention.

[0023] The fourth aspect of the present invention discloses a computer storage medium storing computer instructions, which, when invoked, are used to execute some or all of the steps in the outbound contact intention prediction method based on call behavior temporal attenuation disclosed in the first aspect of the present invention.

[0024] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: This invention obtains customer call behavior and contact event sequences from outbound call logs and extracts behavioral feature parameters. It then constructs a hybrid intention decay model that explicitly displays the customer's intention to contact them, combining individual differences in the contact event sequences with group trends in the behavioral feature table. This hybrid intention decay model is used to calculate the outbound contact intention score of the target customer under their contact event sequence. This enables accurate outbound contact intention scoring based on individual and group intention decay trends, improving the rationality and reach of outbound contact strategies, and reducing the impact of neglecting customer contact intention on the efficiency of non-performing asset disposal. This is particularly significant in the internet finance sector, where compliance is emphasized, as it significantly improves the efficiency and compliance of outbound call operations. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a flowchart illustrating a method for predicting outbound contact intention based on call behavior temporal decay, as disclosed in an embodiment of the present invention.

[0027] Figure 2 This is a schematic diagram of the structure of an outbound call contact intention prediction system based on call behavior temporal decay disclosed in an embodiment of the present invention.

[0028] Figure 3 This is a schematic diagram of another outbound call contact intention prediction system based on call behavior temporal decay disclosed in an embodiment of the present invention. Detailed Implementation

[0029] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0031] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0032] This invention discloses a method and system for predicting outbound contact intention based on the temporal decay of call behavior. It obtains customer call behavior and contact event sequences from outbound call logs and extracts behavioral feature parameters. Combining individual customer differences in the contact event sequences with group trends in the behavioral feature table, a hybrid intention decay model is constructed to explicitly display the customer's contact intention decay process. This hybrid intention decay model is used to calculate the target customer's outbound contact intention score under their contact event sequence. This enables accurate outbound contact intention scoring based on individual and group intention decay trends, improving the rationality and reach of outbound contact strategies, and reducing the impact of neglecting customer contact intention on the efficiency of non-performing asset disposal. This is particularly significant in the compliance-intensive scenarios of internet finance, significantly improving the efficiency and compliance of outbound call work. The following sections provide detailed explanations.

[0033] Example 1 Please see Figure 1 , Figure 1 This is a flowchart illustrating a method for predicting outbound call contact intention based on call behavior temporal attenuation, as disclosed in an embodiment of the present invention. Figure 1 The described outbound call contact intention prediction method based on call behavior temporal decay can be applied to data processing systems / data processing equipment / data processing servers (wherein, the server includes local processing servers or cloud processing servers). For example... Figure 1 As shown, the outbound call contact intention prediction method based on call behavior temporal decay can include the following operations: 101. Obtain customer call behavior through outbound call logs and generate a contact event sequence for each customer's call behavior.

[0034] Optionally, the following preprocessing steps may be included before performing the above steps: Outbound call logs are collected and cleaned and standardized. The outbound call logs include customer identifiers, time sequence identifiers, and fields that record the call behavior. The data cleaning process removes valid outbound calls from the outbound call logs while retaining invalid outbound call data. The standardization process unifies the expression of the call behavior by the fields.

[0035] Optionally, data cleaning of outbound call logs can be performed using Spark to read the raw log content (e.g., in Parquet / CSV format).

[0036] Among them, the filtering of valid outbound calls is mainly used to exclude outbound call test and internal number outbound call data, while the corresponding invalid outbound call data refers to outbound call data for debtor customers who have failed to complete repayment follow-up processing. Since debtor customers have a decreased willingness to make outbound calls, this invalid outbound call data can be used as the basic data for analyzing the relationship between the decreased willingness of debtor customers to make outbound calls and outbound call behavior, thereby improving the efficiency and practicality of outbound repayment follow-up strategies.

[0037] Typically, the raw logs of outbound calls contain different fields representing the same call behavior, so it is necessary to standardize the fields of the raw logs.

[0038] Optionally, call behaviors include outbound call interruption, recall delay, answer interval response, cumulative outbound calls, outbound call rejection, and outbound call complaints. Outbound call interruption refers to the call behavior of a customer hanging up after answering an outbound call. Recall delay refers to the time delay between the current outbound call and the previous outbound call for the same customer. Answer interval response refers to the time delay between the last outbound call being answered and the current outbound call. Cumulative outbound calls refer to the cumulative number of outbound calls made to the customer, including answered, interrupted, missed, or other already made outbound calls. Outbound call rejection refers to the behavior of a customer being rejected within a certain period of time. The number of outbound call interruptions during the period is typically calculated using a sliding window to statistically analyze the behavioral trends of customers associated with outbound call rejections and cumulative outbound calls in the preprocessed outbound call logs within the period. Outbound call complaints refer to customers' complaints and blacklisting of outbound call numbers, which are usually used to determine customer fatigue with outbound calls and to determine the upper limit of the total number of outbound calls for different customers. The above call behaviors are all recorded in the original outbound call logs with corresponding time-series parameters such as connection time, hang-up time, customer ID, and the number of outbound calls, which can be used for subsequent behavioral feature extraction and intention decay analysis.

[0039] Optionally, the contact event sequence length is the number of outbound calls corresponding to the customer's call behavior. The contact event sequence corresponds to the target customer and is used to characterize the target customer's behavioral characteristics under the length of outbound call count. It can not only be used to analyze the willingness of individual customers to show individual differences, but also to statistically analyze the willingness change trend of customer groups, thereby improving the accuracy of willingness decay prediction and assisting in the updating and optimization of outbound call strategies.

[0040] Specifically, the outbound call log data comes from the raw logs stored in the system. Each record contains fields such as: customer ID, outbound call initiation time, call end time, call behavior status (connected / not connected / rejected / interrupted / busy), whether the call was connected (represented by 0 / 1), and call duration. The interruption threshold can be set. For example, the threshold can be defined by a 10-second interval between the connection time and the interruption time. That is, hanging up the call within 10 seconds after the outbound call is connected is considered an interruption. This is stored in the outbound call log in the form of fields.

[0041] 102. Based on the preset feature extraction algorithm, extract the behavioral feature parameters corresponding to each contact event sequence and construct a behavioral feature table.

[0042] Optionally, the feature extraction algorithm sorts all outbound call records corresponding to each customer by time, thereby performing feature engineering on the behavioral feature parameters of the contact event sequence according to feature type. For example, feature types include interruption mode, repeat call delay, response behavior, recent trend, and contact fatigue, etc. The behavioral feature parameters correspond to the behavioral parameters under that feature type. For example, the cumulative number of outbound call interruptions is extracted from the interruption mode. It can be understood that the contact event sequence is used to analyze individual customers, while the behavioral feature table is used to analyze customer groups, respectively studying the nonlinear relationship between call behavior and outbound contact intention attenuation.

[0043] Optionally, the behavioral feature table can be written to ClickHouse by Spark according to customer identifier and time sequence identifier (outbound call event granularity) for storage, so as to facilitate subsequent modeling query calls and model training.

[0044] Optionally, the behavioral characteristic parameters include interruption parameters, repeat call interval parameters, outbound call frequency parameters, historical rejection parameters, and complaint parameters; The interruption parameter represents the cumulative number of call interruptions within the contact event sequence. The repeat call interval parameter is the time interval between the first and second outbound calls within the contact event sequence. The outbound call frequency parameter is the frequency of outbound calls within the contact period of the contact event sequence. The historical rejection parameter is the rejection rate of the customer within the contact period of the contact event sequence, which is the ratio of the number of rejections to the total number of outbound calls. Rejection specifically includes outbound call rejection and / or missed calls. The complaint parameter is the status value corresponding to whether the customer has been blacklisted within the contact event sequence. The complaint parameter takes a value of 0 or 1, where 0 indicates that the outbound number has not been blacklisted and can be called again, while 1 indicates that the outbound number has been blacklisted and cannot be called again. The generation of the contact event sequence and the construction of the behavioral feature table are both data preparations for the intention decay analysis.

[0045] 103. Determine the intention decay parameters based on the behavioral feature table and the contact event sequence, and construct a hybrid intention decay model.

[0046] Optionally, the willingness decay parameters include individual initial willingness, interruption sensitivity coefficient, delay recovery rate, and complaint interception index; The individual initial intention is used to characterize the differences in individual intentions of any customer in the initial contact in a sequence of contact events; The interruption sensitivity coefficient is used to characterize the intensity of the impact of each increase in outbound call interruption on the decline in contact willingness; The delayed recovery rate is used to characterize the ability of any customer to resume contact during the outbound call interval; The complaint interception metric is used to characterize any customer's willingness to be contacted regarding a complaint or non-complaint. The individual's initial intention and the interruption sensitivity coefficient are associated with any customer's contact event sequence, which constitutes the driving attenuation term of the outbound call interruption behavior pattern. The delayed recovery rate is associated with the sequence of contact events for any customer, forming a delayed recovery factor term in the response interval response behavior pattern. Based on the driving attenuation term and the delay recovery factor term, a nonlinear model is performed to explicitly model the nonlinear relationship between the customer's outbound call interruption behavior pattern and the response interval response behavior pattern in the contact event sequence.

[0047] Specifically, the hybrid intention decay model consists of multiple models corresponding to the intention decay parameters associated with different behavioral feature dimensions. By dynamically capturing the intention decay features associated with call behavior, it determines the impact of call behavior on outbound contact intention. Through interpretable, computable, and deployable mathematical expressions, it explicitly demonstrates the customer's contact intention decay process by jointly modeling the call interruption pattern and repeat call response delay, thereby improving the outbound contact reach rate.

[0048] 104. Calculate the outbound contact willingness score of the target customer based on the contact event sequence of the target customer and the hybrid willingness decay model.

[0049] As can be seen, the above-mentioned embodiments of the invention obtain customer call behavior and contact event sequences through outbound call logs and extract behavioral feature parameters. Combining the individual differences of customers in the contact event sequences with the group trends in the behavioral feature table, a hybrid intention decay model is constructed to explicitly show the customer's contact intention decay process. The hybrid intention decay model is used to calculate the outbound contact intention score of the target customer under its contact event sequence. This enables accurate outbound contact intention scores based on individual intention decay trends and group intention decay trends, improves the rationality and reach of outbound contact strategies, and reduces the impact of ignoring customer contact intentions on the efficiency of non-performing asset disposal. Especially in the Internet finance sector, where compliance is emphasized, it significantly improves the efficiency and compliance of outbound work.

[0050] As an optional embodiment, the step of constructing the hybrid intention decay model in the above steps includes: A loss-of-contact risk model is constructed based on survival analysis. The behavioral feature table and the contact event sequence of any customer are obtained and input into the loss-of-contact risk model to obtain the loss-of-contact risk value of the corresponding customer to distinguish risky customers. The loss-of-contact risk model is trained by the behavioral feature table and the contact event sequence. Optionally, the lost contact risk model can use the Cox model for survival analysis, where the lost contact risk is associated with the willingness to make outbound calls. It can be understood that the higher the lost contact risk, the lower the willingness to make outbound calls. Specifically, it is used to reveal the relationship between outbound call interruption and lost contact risk. On the one hand, an increase in the number of outbound call interruptions will lead to an increase in lost contact risk. On the other hand, an increase in the number of outbound call interruptions will also reduce the willingness to make outbound calls. The lost contact risk model can be used to screen risky customers and set different outbound call strategies.

[0051] The behavioral feature table and the contact event sequence of any customer are obtained as input. Based on Bayesian exponential decay fitting, the individual initial intention of the intention decay parameter and the repeat call threshold corresponding to the individual initial intention are obtained. Among them, the Bayesian fitting estimation examines the posterior distribution of the intention decay parameter of an individual's initial intention to improve the accuracy of the intention decay parameter fitting estimation, while the call-back threshold can be used as a personalized call-back threshold set for the target customer. When the intention is less than the call-back threshold, the outbound calls to the target customer are suspended.

[0052] The behavioral feature table and the aggregated contact event sequence are used as inputs, and the interruption sensitivity coefficient and delay recovery rate of the intention decay parameter are obtained by sliding window estimation fitting. Optionally, the interruption sensitivity coefficient and delay recovery rate can be fitted nonlinearly using the logistic function.

[0053] The system obtains a sequence of contact events for any customer as input, and dynamically captures the decay of contact intention in the sequence of contact events based on a neural network prediction model to predict the customer's willingness to call back at the last moment. The neural network prediction model is trained using the behavioral feature table and the sequence of contact events. Optionally, the neural network prediction model can be an LSTM model or a Transformer model, where the behavioral feature table can be used as the training set for the neural network prediction model, and the customer's contact event sequence can be used as the test set to verify the accuracy of the contact willingness score predicted at the t-th outbound call.

[0054] The system obtains a contact event sequence of any customer as input, and uses the LightGBM model to perform auxiliary estimation of the repeat contact intention score on the behavioral feature parameters in the contact event sequence, so as to help correct the repeat contact intention score. The LightGBM model is trained by the behavioral feature table, the contact event sequence and the labeled repeat contact intention score.

[0055] As can be seen, through the above optional embodiments, by training multiple models based on behavioral features and contact events, and by determining the intention decay parameter through fitting estimation, scientific support is provided for determining the dynamic intention decay data of customers based on the intention decay parameter accurately expressing the relationship between customer call behavior characteristics and outbound call intention. The personalization and reliability of individual customer sequence analysis are improved by calibrating customer group data, and the targeting and continuity of outbound call intention prediction are improved by the corresponding model. This provides an accurate score as a basis for judgment for subsequent outbound call strategies, reduces the risk of bias in outbound call strategies due to ignoring the relationship between customer call behavior and contact intention decay, and improves the reach rate of outbound calls.

[0056] As an optional embodiment, the step of constructing the hybrid intention decay model in the above steps further includes: The intention decay function is modeled based on the lost contact risk model and the Bayesian exponential decay fitting. The intention decay function is then probabilistically modeled and nondeterministically quantified based on the Bayesian exponential decay fitting. Finally, the customer's outbound contact intention score is predicted based on the intention decay function using the LightGBM model and a neural network prediction model, resulting in the hybrid intention decay model.

[0057] As can be seen, through the above optional embodiments, the dynamic decline in customer willingness to contact is modeled based on outbound call data using a loss-of-contact risk model and Bayesian exponential decay fitting. Furthermore, Bayesian exponential decay fitting is used to perform more accurate probability modeling based on customer group behavior characteristics and individual customer time-series characteristics, while quantifying the uncertainty of the probability component. This improves the granularity and interpretability of the dynamic willingness decline process, thereby providing a comprehensive indicator of outbound contact willingness scoring for outbound call strategies. This reduces the risk of scoring distortion caused by improper calculation of feature dimensions, ultimately achieving an interpretable, computable, and deployable transparent logical expression for outbound call strategies, improving the efficiency and compliance of non-performing asset disposal work.

[0058] As an optional embodiment, the calculation expression for the target customer's outbound contact willingness score in the above steps is: ; ; ; in, This indicates that the hybrid intention decay model calculates the target customer's initial contact intention score. This indicates that the auxiliary correction model predicts and estimates the target customer's second contact intention score. This indicates the target customer's willingness to be contacted via outbound calls. This represents the initial individual intention of the target customer. Represents the natural constant. This represents the interruption sensitivity coefficient. This indicates the interruption parameters for the target customer. Indicates the latency recovery rate of the target customer. This represents the repeat call interval parameter for the target customer. This indicates the metrics for intercepting complaints from target customers. This represents the intention estimation function. This parameter represents the outbound call frequency within a period d of the target customer contact event sequence. This indicates the historical rejection parameters of the target customer. The score weights represent the predictions of declining willingness. Let i represent the sequence number of the target customer i, and t represent the t-th outbound call in the contact event sequence of the target customer.

[0059] Specifically, as shown in the first contact willingness scoring formula, the individual's initial willingness and interruption sensitivity coefficient constitute a driving attenuation term based on the interruption parameters in the customer contact event sequence, used to analyze the correlation between outbound call interruption behavior patterns and the attenuation process of outbound contact willingness. The delay recovery rate and repeat call interval parameters constitute a delay recovery factor term, used to analyze the correlation between the response interval response behavior model and the outbound contact willingness recovery process. Correspondingly, the driving attenuation term and the delay recovery factor term are based on an exponential relationship with the natural constant e as the base, used to analyze the nonlinear relationship of customer willingness attenuation under the combination of outbound call interruption behavior patterns and response interval response behavior patterns, thereby explicitly modeling it. It can be understood that when the complaint interception index is 1, that is, the customer has complained and blacklisted the outbound number, the contact willingness is 0. The scoring weight of the willingness attenuation prediction is an empirical value, which is set accordingly based on the update of the hybrid willingness attenuation model.

[0060] As can be seen, through the above optional embodiments, the intention decay function analyzes the impact of the intention decay of the next outbound call corresponding to the customer's contact event sequence on the intention to make an outbound call, and outputs a first contact intention score. An auxiliary correction model trained on customer group behavioral characteristics predicts a second contact intention score for the next outbound call based on the target customer's contact event sequence. Combining the two intention scores determines the target customer's outbound contact intention score. Based on accurate intention decay prediction, the granularity and interpretability of the outbound contact intention score are improved through individual differences and group decay trends. This provides comprehensive indicators for subsequent outbound call strategies, reduces the risk of score distortion due to improper calculation of feature dimensions, and ultimately achieves an interpretable, calculable, and deployable transparent logical expression for outbound call strategies, improving the efficiency and compliance of non-performing asset disposal work.

[0061] In one specific implementation scheme, an outbound call contact willingness scoring system based on call behavior time-series decay is implemented based on the technical solution of this embodiment. It mainly collects multi-dimensional call behavior time-series data of customers from outbound call logs, and uses the willingness decay model proposed in this application to fuse and analyze the time-series data of target customers, ultimately generating an intuitive and personalized contact willingness score. Specifically, in the application scenarios of this system, the implementation process of the system is illustrated through the following typical outbound call contact scenarios: Scenario 1: Customer A is a highly sensitive customer with rapidly declining willingness to make outbound calls. The first outbound call connects but hangs up after 5 seconds (1 interruption); two hours later, the outbound call hangs up again (2 interruptions). The system calculates the input interruption parameter as 2 and the repeat call interval parameter as 2. Substituting these into the willingness decline model, the calculated willingness score for the next outbound call is 0.12. Based on the score, the outbound call strategy is set to suspend outbound calls for a certain period (e.g., within 48 hours, the specific duration of which is determined based on the delay recovery rate) to avoid the risk of losing contact. After determining this period, repeat calls are made to improve the outbound call reach rate.

[0062] Scenario 2: Customer B is a low-sensitivity customer with slowly declining willingness. In Customer B's contact event sequence, the first three outbound calls were missed (not outbound call interruptions), while the fourth outbound call was successfully connected and involved communication about non-performing asset disposal. The system calculates the input interruption parameter as 0, but the input repeat call interval parameter continuously increases (the sequence length increases), and the output outbound contact willingness score is always greater than 0.4. Based on the score, the outbound call strategy is set to continuously make outbound calls until the customer is reached within the willingness recovery period (e.g., within 24-48 hours). This willingness recovery period is also calculated based on the hybrid willingness decline model. By distinguishing between missed calls and interruptions through the model, the granularity and accuracy of willingness prediction are improved.

[0063] Specifically, at the system architecture level, the components of this system architecture include: Model service components are used for low-latency prediction of customers' outbound contact intentions; The message queue component is used to decouple outbound call requests from outbound call decisions; The strategy engine component is used to couple the model with outbound call compliance rules. The testing platform component is used to evaluate the effectiveness and monitor in real time the outbound calling strategy based on outbound contact willingness prediction. The model update component is used to update model parameters in real time based on log data and model output; Among them, the model service component, message queue component, strategy engine component, verification platform component, and model update component can all be deployed in containers using Docker, and can achieve elastic scaling in conjunction with Kubernetes.

[0064] Furthermore, the model service component accelerates inference by exporting models to ONNX format and deploying ONNX models using FastAPI.

[0065] Furthermore, the message queue component integrates with Kafka for event-driven scheduling and customer service, the steps of which include: invoking the model service; executing outbound call strategies; and sending instructions to the outbound call system.

[0066] Furthermore, the platform components include A / B testing and feedback loops. The A / B testing configuration uses MLflow to track and compare the fixed outbound calling strategy with the outbound calling strategy after the intention decay prediction.

[0067] Furthermore, the model update component uses River to learn and update the model online. The online model can serve as a supplementary model to the main model (i.e., the hybrid intention decay model) in the model service component, enabling rapid adaptation to short-term changes.

[0068] Specifically, the system architecture also includes monitoring and compliance components.

[0069] The monitoring component uses the IT monitoring and management tool Prometheus and the data visualization and analysis tool Grafana. First, it connects to the model service of the system framework to define monitoring metrics. Outbound call and complaint events are reported through the business system. The Grafana dashboard is configured with visualization metrics and alarm rules. The visualization metrics include model performance (such as the distribution of contact intention scores and the predicted query rate and latency per second), business performance (such as outbound call reach success rate and complaint rate), and alarm rules (such as the complaint rate abnormally increasing beyond a preset threshold).

[0070] Among them, the compliance component uses differential privacy and federated learning for privacy protection modeling and multi-institution writing modeling. Differential privacy performs noise enhancement and gradient pruning during local training of the model service. Federated learning can use the Flower framework to deploy clients locally in various institutions in joint modeling with multiple parties (such as banks, debt collection companies, etc.). It is worth noting that the above compliance components enable local institutions to meet the requirements for de-identified personal information protection, and multi-institution joint modeling ensures that the original logs of outbound calls required in the system architecture of this application are kept locally, and multiple institutions only exchange model parameters, further improving the compliance of non-performing asset disposal work.

[0071] Example 2 Please see Figure 2 , Figure 2 This is a schematic diagram of an outbound call contact intention prediction system based on call behavior temporal attenuation, as disclosed in an embodiment of the present invention. Figure 2 The described outbound contact intention prediction system based on call behavior temporal decay can be applied to data processing systems / data processing equipment / data processing servers (wherein, the server includes local processing servers or cloud processing servers). For example... Figure 2 As shown, the outbound contact intention prediction system based on call behavior temporal decay may include: The acquisition module 201 is used to acquire customer call behavior through outbound call logs and generate a contact event sequence for each customer's call behavior.

[0072] The extraction module 202 is used to extract behavioral feature parameters corresponding to each contact event sequence based on a preset feature extraction algorithm, and construct a behavioral feature table.

[0073] The determination module 203 is used to determine the intention decay parameters based on the behavioral feature table and the contact event sequence, and to construct a hybrid intention decay model.

[0074] The scoring module 204 is used to calculate the outbound contact willingness score of the target customer based on the contact event sequence of the target customer and the hybrid willingness decay model.

[0075] As can be seen, the above-mentioned embodiments of the invention obtain customer call behavior and contact event sequences through outbound call logs and extract behavioral feature parameters. Combining the individual differences of customers in the contact event sequences with the group trends in the behavioral feature table, a hybrid intention decay model is constructed to explicitly show the customer's contact intention decay process. The hybrid intention decay model is used to calculate the outbound contact intention score of the target customer under its contact event sequence. This enables accurate outbound contact intention scores based on individual intention decay trends and group intention decay trends, improves the rationality and reach of outbound contact strategies, and reduces the impact of ignoring customer contact intentions on the efficiency of non-performing asset disposal. Especially in the Internet finance sector, where compliance is emphasized, it significantly improves the efficiency and compliance of outbound work.

[0076] As an optional implementation, before obtaining the customer's call behavior through outbound call logs, the following steps are included: Outbound call logs are collected and cleaned and standardized. The outbound call logs include customer identifiers, time sequence identifiers, and fields that record the call behavior. The data cleaning process removes valid outbound calls from the outbound call logs while retaining invalid outbound call data. The standardization process unifies the expression of the call behavior by the fields.

[0077] As can be seen from the above embodiments, valid outbound calls are used to exclude outbound call testing and outbound call data from internal numbers, while the corresponding invalid outbound call data refers to outbound call data for which debtor customers have failed to complete repayment follow-up processing. Since debtor customers have a decreased willingness to make outbound calls, this invalid outbound call data can be used as basic data to analyze the relationship between the decreased willingness of debtor customers to make outbound calls and outbound call behavior, thereby improving the efficiency and practicality of outbound repayment follow-up strategies.

[0078] As an optional implementation, the call behavior includes outbound call interruption, call retry delay, answer interval response, cumulative outbound calls, outbound call rejection, and outbound call complaint; The length of the contact event sequence is the number of outbound calls made by the customer for the corresponding call behavior.

[0079] As can be seen, through the above embodiments, by further defining the call behavior and contact event sequence, it can not only be used to analyze the intentions of individual customers to show individual differences, but also to statistically analyze the trend of intention changes in customer groups, thereby improving the accuracy of intention decay prediction and assisting in the updating and optimization of outbound calling strategies.

[0080] As an optional implementation, the behavioral characteristic parameters include interruption parameters, repeat call interval parameters, outbound call frequency parameters, historical rejection parameters, and complaint parameters; The interruption parameter is the cumulative number of call actions corresponding to the customer's call interruption within and outside the contact event sequence; The repeat call interval parameter is the time interval between the outbound call at the first moment and the outbound call at the second moment within the contact event sequence; The outbound call frequency parameter is the frequency of outbound call behavior within the contact cycle of the contact event sequence; The historical rejection parameter is the rejection rate of a customer within the contact period of a contact event sequence. The rejection rate is the ratio between the number of rejection behaviors and the total number of outbound calls. The rejection behaviors specifically include outbound call rejection and / or outbound call non-answer. The complaint parameter refers to the status value corresponding to whether the customer has complained to the blacklist within the contact event sequence.

[0081] As can be seen, through the above embodiments, by further defining the behavioral feature parameters, the model application requirements corresponding to the multiple intention decay parameters associated with different behavioral feature dimensions in the hybrid intention decay model are met. By dynamically capturing the intention decay features associated with call behavior, the impact of call behavior on outbound contact intention is determined. Through interpretable, computable, and deployable mathematical expressions, the joint modeling of call interruption patterns and repeat call response delay explicitly demonstrates the customer's contact intention decay process, thereby improving the outbound contact reach rate.

[0082] As an optional implementation, the willingness decay parameters include individual initial willingness, interruption sensitivity coefficient, delay recovery rate, and complaint interception index; The individual initial intention is used to characterize the differences in individual intentions of any customer in the initial contact in a sequence of contact events; The interruption sensitivity coefficient is used to characterize the intensity of the impact of each increase in outbound call interruption on the decline in contact willingness; The delayed recovery rate is used to characterize the ability of any customer to resume contact during the outbound call interval; The complaint interception metric is used to characterize any customer's willingness to be contacted regarding a complaint or non-complaint. The individual's initial intention and the interruption sensitivity coefficient are associated with any customer's contact event sequence, which constitutes the driving attenuation term of the outbound call interruption behavior pattern. The delayed recovery rate is associated with the sequence of contact events for any customer, forming a delayed recovery factor term in the response interval response behavior pattern. Based on the driving attenuation term and the delay recovery factor term, a nonlinear model is performed to explicitly model the nonlinear relationship between the customer's outbound call interruption behavior pattern and the response interval response behavior pattern in the contact event sequence.

[0083] As can be seen, through the above embodiments, by further defining the intention decay parameter, the behavioral feature parameter is associated with the intention decay parameter. The impact of call behavior on outbound contact intention is determined by dynamically capturing the intention decay characteristics associated with call behavior. The customer's contact intention decay process is explicitly displayed by the joint modeling of call interruption mode and repeat call response delay through interpretable, computable and deployable mathematical expressions, thereby improving the outbound contact reach rate.

[0084] As an optional implementation, in a second aspect of the invention, constructing the hybrid intention decay model includes: A loss-of-contact risk model is constructed based on survival analysis. The behavioral feature table and the contact event sequence of any customer are obtained and input into the loss-of-contact risk model to obtain the loss-of-contact risk value of the corresponding customer to distinguish risky customers. The loss-of-contact risk model is trained by the behavioral feature table and the contact event sequence. The behavioral feature table and the contact event sequence of any customer are obtained as input. Based on Bayesian exponential decay fitting, the individual initial intention of the intention decay parameter and the repeat call threshold corresponding to the individual initial intention are obtained. The behavioral feature table and the aggregated contact event sequence are used as inputs, and the interruption sensitivity coefficient and delay recovery rate of the intention decay parameter are obtained by sliding window estimation fitting. The system obtains a sequence of contact events for any customer as input, and dynamically captures the decay of contact intention in the sequence of contact events based on a neural network prediction model to predict the customer's willingness to call back at the last moment. The neural network prediction model is trained using the behavioral feature table and the sequence of contact events. The system obtains a contact event sequence of any customer as input, and uses the LightGBM model to perform auxiliary estimation of the repeat contact intention score on the behavioral feature parameters in the contact event sequence, so as to help correct the repeat contact intention score. The LightGBM model is trained by the behavioral feature table, the contact event sequence and the labeled repeat contact intention score.

[0085] As can be seen, through the above embodiments, multiple models are trained based on behavioral features and contact events, and the willingness decay parameter is determined by fitting estimation. Thus, based on the fact that the willingness decay parameter accurately expresses the relationship between customer call behavior characteristics and outbound call willingness, it provides scientific support for determining dynamic willingness decay data of customers. It improves the personalization and reliability of individual customer sequence analysis by calibrating customer group data, and improves the targeting and continuity of outbound call willingness prediction by corresponding models. It provides accurate scores as a basis for judgment for subsequent outbound call strategies, reduces the risk of bias in outbound call strategies due to ignoring the relationship between customer call behavior and contact willingness decay, and improves the reach rate of outbound calls.

[0086] As an optional implementation, the construction of the hybrid intention decay model further includes: The intention decay function is modeled based on the lost contact risk model and the Bayesian exponential decay fitting. The intention decay function is then probabilistically modeled and nondeterministically quantified based on the Bayesian exponential decay fitting. Finally, the customer's outbound contact intention score is predicted based on the intention decay function using the LightGBM model and a neural network prediction model, resulting in the hybrid intention decay model.

[0087] As can be seen, through the above embodiments, the dynamic decline in customer willingness to make outbound calls is modeled based on outbound call data using a loss-of-contact risk model and Bayesian exponential decay fitting. Furthermore, Bayesian exponential decay fitting is used to perform more accurate probability modeling based on the behavioral characteristics of customer groups and the temporal characteristics of individual customers, while quantifying the uncertainty of the probability component. This improves the granularity and interpretability of the dynamic willingness decline process, thereby providing a comprehensive indicator of outbound call contact willingness scoring for outbound call strategies. This reduces the risk of scoring distortion caused by improper calculation of feature dimensions, ultimately achieving an interpretable, computable, and deployable transparent logical expression for outbound call strategies, improving the efficiency and compliance of non-performing asset disposal work.

[0088] As an optional implementation, the formula for calculating the target customer's outbound contact willingness score is: ; ; ; in, This indicates that the hybrid intention decay model calculates the target customer's initial contact intention score. This indicates that the auxiliary correction model predicts and estimates the target customer's second contact intention score. This indicates the target customer's willingness to be contacted via outbound calls. This represents the initial individual intention of the target customer. Represents the natural constant. This represents the interruption sensitivity coefficient. This indicates the interruption parameters for the target customer. Indicates the latency recovery rate of the target customer. This represents the repeat call interval parameter for the target customer. This indicates the metrics for intercepting complaints from target customers. This represents the intention estimation function. This parameter represents the outbound call frequency within a period d of the target customer contact event sequence. This indicates the historical rejection parameters of the target customer. The score weights represent the predictions of declining willingness. Let i represent the sequence number of the target customer i, and t represent the t-th outbound call in the contact event sequence of the target customer.

[0089] As can be seen from the above embodiments, the intention decay function analyzes the impact of the intention decay of the next outbound call corresponding to the customer's contact event sequence on the intention to make outbound calls, and outputs a first contact intention score. An auxiliary correction model trained on customer group behavioral characteristics predicts a second contact intention score for the next outbound call based on the target customer's contact event sequence. Combining the two intention scores determines the target customer's outbound call intention score. Based on accurate intention decay prediction, the granularity and interpretability of the outbound call intention score are improved through individual differences and group decay trends. This provides comprehensive indicators for subsequent outbound call strategies, reduces the risk of score distortion due to improper calculation of feature dimensions, and ultimately achieves a transparent logical expression for outbound call strategies that is interpretable, calculable, and deployable, improving the efficiency and compliance of non-performing asset disposal work.

[0090] Example 3 Please see Figure 3 , Figure 3 This is another outbound call contact intention prediction system based on call behavior temporal decay disclosed in the embodiments of the present invention. Figure 3 The described outbound contact intention prediction system based on call behavior temporal decay is applied in a data processing system / data processing equipment / data processing server (wherein, the server includes a local processing server or a cloud processing server). Figure 3 As shown, the outbound contact intention prediction system based on call behavior temporal decay may include: Memory 301 storing executable program code; Processor 302 coupled to memory 301; The processor 302 calls the executable program code stored in the memory 301 to execute the steps of the outbound contact intention prediction method based on call behavior temporal decay described in Embodiment 1.

[0091] Example 4 This invention discloses a computer read storage medium that stores a computer program for electronic data interchange, wherein the computer program causes a computer to execute the steps of the outbound contact intention prediction method based on call behavior temporal decay described in Embodiment 1.

[0092] Example 5 This invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to perform the steps of the outbound contact intention prediction method based on call behavior temporal decay described in Embodiment 1.

[0093] The foregoing has described specific embodiments of this specification; other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than those shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily have to follow the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0094] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0095] For ease of description, the above devices are described in terms of function, divided into various units. Of course, in implementing this specification, the functions of each unit can be implemented in one or more software and / or hardware components.

[0096] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, the embodiments of this specification can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the embodiments of this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0097] This specification is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this specification. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0098] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0099] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0100] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0101] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0102] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0103] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0104] This specification can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This specification can also be practiced in distributed computing environments, where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0105] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0106] Finally, it should be noted that the outbound contact intention prediction method and system based on call behavior temporal attenuation disclosed in the embodiments of the present invention are merely preferred embodiments of the present invention, and are only used to illustrate the technical solutions of the present invention, not 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; and these 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 call behavior timing-based attenuation-based outbound contact willingness prediction method, characterized in that, The method comprises: obtaining the call behavior of the customer through the outbound call log, and generating a contact event sequence for each customer's call behavior; based on the preset feature extraction algorithm, extracting the behavior feature parameters corresponding to each contact event sequence, and constructing a behavior feature table; determining the willingness decay parameter according to the behavior feature table and the contact event sequence, and constructing a mixed willingness decay model; According to the contact event sequence of the target customer and the mixed willingness decay model, the outbound contact willingness score of the target customer is calculated. 2.The call behavior timing-based attenuation-based outbound contact willingness prediction method according to claim 1, characterized in that, Before obtaining the call behavior of the customer through the outbound call log, it comprises: Collecting outbound call logs for data cleaning and standardization processing, wherein the outbound call log includes customer identification, time sequence identification and fields recording the call behavior; The data cleaning eliminates the valid outbound calls in the outbound call log to retain invalid outbound call data; The standardization processing unifies the expression of the fields to the call behavior. 3.The call behavior timing-based attenuation-based outbound contact willingness prediction method according to claim 1, characterized in that, The call behavior includes outbound call interruption, repeated call delay, response interval response, cumulative outbound call, outbound call rejection and outbound call complaint; The length of the contact event sequence is the number of outbound calls corresponding to the call behavior of the customer. 4.The call behavior timing-based attenuation-based outbound contact willingness prediction method according to claim 1, characterized in that, The behavior feature parameters include interruption parameters, repeated call interval parameters, outbound call frequency parameters, historical rejection parameters and complaint parameters; The interruption parameter is the cumulative number of call behaviors corresponding to the customer's outbound call interruption in the contact event sequence; The repeated call interval parameter is the time interval between the first outbound call and the second outbound call in the contact event sequence; The outbound call frequency parameter is the frequency of the outbound call behavior in the contact period of the contact event sequence; The historical rejection parameter is the rejection rate of the customer in the contact period of the contact event sequence, and the rejection rate is the ratio between the number of rejection behaviors and the total number of outbound calls, and the rejection behavior is specifically outbound call rejection and / or outbound call non-connection; The complaint parameter is a state value corresponding to whether the customer complains about the blacklist in the contact event sequence. 5.The call behavior timing-based attenuation-based outbound contact willingness prediction method according to claim 1, characterized in that, The willingness decay parameter includes individual initial willingness, interruption sensitivity coefficient, delay recovery rate and complaint interception index; The individual initial willingness is used to represent the individual willingness difference of any customer in the initial contact of the contact event sequence; The interruption sensitivity coefficient is used to represent the influence intensity of the increase of each outbound call interruption on the contact willingness decay; The delay recovery rate is used to represent the contact willingness recovery ability of any customer during the outbound call interval; The complaint interception index is used to represent the contact willingness of any customer in the complaint / non-complaint; The individual initial willingness and the interruption sensitivity coefficient are associated with the contact event sequence of any customer to form a driving decay term of the outbound call interruption behavior mode; The delay recovery rate is associated with the contact event sequence of any customer to form a delay recovery factor term of the response interval response behavior mode; Based on the driving decay term and the delay recovery factor term, the nonlinear relationship between the customer's outbound call interruption behavior mode and the response interval response behavior mode in the contact event sequence is displayed modeling. 6.The call behavior timing-based attenuation-based outbound contact willingness prediction method according to claim 5, characterized in that, The construction of the mixed willingness decay model comprises: constructing a disconnection risk model based on survival analysis, inputting the behavior feature table and the contact event sequence of any customer into the disconnection risk model to obtain a disconnection risk value of the corresponding customer to distinguish risk customers, wherein the disconnection risk model is obtained by training the behavior feature table and the contact event sequence; obtaining the behavior feature table and the contact event sequence of any customer as input, and obtaining the individual initial willingness of the willingness decay parameter and the corresponding re-call threshold value of the individual initial willingness based on Bayesian exponential decay fitting; obtaining the behavior feature table and the aggregated contact event sequence as input, and obtaining the interruption sensitivity coefficient and the delay recovery rate of the willingness decay parameter based on sliding window estimation fitting; obtaining the contact event sequence of any customer as input, and dynamically capturing the contact willingness decay in the contact event sequence based on a neural network prediction model to predict the re-call contact willingness score of the customer at the last moment, wherein the neural network prediction model is obtained by training the behavior feature table and the contact event sequence; obtaining the contact event sequence of any customer as input, and performing auxiliary estimation of the behavior feature parameters in the contact event sequence based on a LightGBM model to assist in correcting the re-call contact willingness score, wherein the LightGBM model is obtained by training the behavior feature table, the contact event sequence, and the labeled re-call contact willingness score.

7. The call behavior timing-based attenuation-based outbound contact willingness prediction method according to claim 6, characterized in that, The construction of the mixed willingness decay model further comprises: modeling the willingness decay function based on the disconnection risk model and the Bayesian exponential decay fitting, probabilistically modeling and non-deterministically quantifying the willingness decay function based on the Bayesian exponential decay fitting, and predicting the outbound contact willingness score of the customer based on the willingness decay function based on the LightGBM model and the neural network prediction model to obtain the mixed willingness decay model. 8.The call behavior timing-based attenuation-based outbound contact willingness prediction method according to claim 1, characterized in that, The calculation expression of the outbound contact willingness score of the target customer is: ; ; ; wherein, represents a first contact willingness score of the target customer calculated by a mixed willingness decay model, represents a second contact willingness score of the target customer estimated by an auxiliary correction model, represents an outbound contact willingness score of the target customer, represents an individual initial willingness of the target customer, represents a natural constant, represents an interruption sensitivity coefficient, represents an interruption parameter of the target customer, represents a delay recovery rate of the target customer, represents a callback interval parameter of the target customer, represents a complaint interception indicator of the target customer, represents a willingness estimation function, represents an outbound frequency parameter within a period d of a contact event sequence of the target customer, represents a historical rejection parameter of the target customer, represents a score weight of the willingness decay prediction, represents a serial number of the target customer i, and t represents the tth outbound call in a contact event sequence of the target customer.

9. A system for predicting outbound contact intention based on call behavior temporal decay, characterized in that, The system comprises: an acquisition module configured to acquire the call behavior of the customer through an outbound log, and generate a contact event sequence for the call behavior of each customer; an extraction module configured to extract the behavior feature parameters corresponding to each contact event sequence based on a preset feature extraction algorithm, and construct a behavior feature table; a determination module configured to determine the willingness decay parameter based on the behavior feature table and the contact event sequence, and construct a mixed willingness decay model; a calculation module configured to calculate the outbound contact willingness score of the target customer based on the contact event sequence of the target customer and the mixed willingness decay model.

10. A system for predicting outbound call contact intention based on call behavior temporal decay, characterized in that, The system comprises: a memory storing executable program codes; a processor coupled with the memory; the processor invokes the executable program codes stored in the memory to execute the outbound contact willingness prediction method based on the call behavior time sequence decay according to any one of claims 1-8.

Images