Predictive outbound calling method, apparatus, device, storage medium and computer program product

Abstract

The application relates to the technical field of telephone outbound call, and discloses a predictive outbound call method, device, equipment, storage medium and computer program product, the method comprising the following steps: based on collected outbound service data, multidimensional feature extraction is carried out in the time dimension, the space dimension and the behavior dimension respectively, target features of the outbound service data in each dimension are obtained; each target feature is weighted and fused through an attention mechanism to obtain a target multidimensional feature; the target multidimensional feature is input into a preset deep neural network model to obtain an outbound call strategy vector; an outbound task list is generated based on a preset dynamic regulation rule and the outbound call strategy vector, and outbound call is carried out according to the outbound task list. Since the application considers the correlation between multidimensional features to obtain the outbound call strategy vector, the efficiency of the predictive outbound call is improved.

Description

Technical Field

[0001] This application relates to the field of outbound calling technology, and in particular to a predictive outbound calling method, apparatus, device, storage medium, and computer program product. Background Technology

[0002] Predictive outbound calling is a call center automation technology based on intelligent algorithms and real-time data. Its core objective is to maximize agent utilization, improve outbound calling efficiency, and reduce operating costs while ensuring customer experience.

[0003] Existing technologies generally use linear statistical models to obtain outbound call strategy vectors, including: number of outbound calls, number priority, agent allocation ratio, predicted connection rate, and agent utilization rate. The outbound call task list is determined by the above outbound call strategy vectors, and outbound calls are made according to the outbound call task list. However, since linear statistical models can only support obtaining the correlation between low-dimensional features to obtain outbound call strategy vectors, the proportion of invalid outbound calls is high, resulting in low efficiency of predictive outbound calls. Summary of the Invention

[0004] The main purpose of this application is to provide a predictive outbound calling method, which aims to solve the technical problem of how to improve the efficiency of predictive outbound calling.

[0005] To achieve the above objectives, this application proposes a predictive outbound calling method, the method comprising:

[0006] Based on the collected outbound call service data, multidimensional feature extraction is performed in the time dimension, spatial dimension and behavioral dimension to obtain the target features of the outbound call service data in each dimension.

[0007] The target features are weighted and fused using an attention mechanism to obtain multidimensional target features;

[0008] The target multidimensional features are input into a preset deep neural network model to obtain the outbound call strategy vector;

[0009] An outbound call task list is generated based on preset dynamic control rules and the outbound call strategy vector, and outbound calls are made according to the outbound call task list.

[0010] In one embodiment, the step of extracting multi-dimensional features from the collected outbound call service data in the time, space, and behavioral dimensions to obtain the target features of the outbound call service data in each dimension includes:

[0011] The collected outbound call service data is feature extracted in the time dimension to obtain at least the time period code and outbound call time interval of the outbound call service data in the time dimension.

[0012] Feature extraction is performed on the outbound call service data in the spatial dimension to obtain at least the agent grouping and number location code of the outbound call service data in the spatial dimension.

[0013] Feature extraction is performed on the outbound call service data in the behavioral dimension to obtain at least the user's historical call connection count and agent-user matching degree in the behavioral dimension of the outbound call service data.

[0014] In one embodiment, the step of weightedly fusing the target features using an attention mechanism to obtain multidimensional target features includes:

[0015] Based on the target features described above, determine the time feature vector, spatial feature vector, and behavioral feature vector;

[0016] The temporal feature vector, the spatial feature vector, and the behavioral feature vector are horizontally concatenated to generate an original feature matrix, and the attention weights corresponding to each target feature are determined based on the attention mechanism and the original feature matrix.

[0017] The target features are weighted and fused based on the attention weights, and the weighted fusion result is normalized to obtain multidimensional target features.

[0018] In one embodiment, the step of horizontally concatenating the temporal feature vector, the spatial feature vector, and the behavioral feature vector to generate the original feature matrix includes:

[0019] The vector length of the time feature vector, the spatial feature vector, and the behavioral feature vector are respectively detected.

[0020] When the length detection result shows that the vector lengths are inconsistent, the time feature vector, the spatial feature vector, and the behavioral feature vector are padded based on a preset fixed length and a preset padding value, respectively.

[0021] The filled time feature vector, spatial feature vector, and behavioral feature vector are horizontally concatenated to generate the original feature matrix.

[0022] In one embodiment, the step of generating an outbound call task list based on preset dynamic control rules and the outbound call strategy vector includes:

[0023] Obtain the predicted connection rate and predicted agent utilization rate from the outbound call strategy vector;

[0024] The number priority is determined based on the preset dynamic control rules and the predicted connection rate, and the seat allocation scheme is determined based on the preset dynamic control rules and the predicted seat utilization rate.

[0025] A list of outbound call tasks is generated based on the number priority and the agent allocation scheme.

[0026] In one embodiment, after the step of making outbound calls according to the outbound call task list, the method further includes:

[0027] Obtain outbound call results, and determine the actual connection rate, actual agent utilization rate, and actual call loss rate based on the outbound call results;

[0028] Incremental samples are determined based on preset learning parameters, the actual connection rate, the actual agent utilization rate, and the actual call loss rate.

[0029] The preset deep neural network model is updated based on the incremental samples to obtain a new preset deep neural network model;

[0030] When a new target multidimensional feature is generated, the new target multidimensional feature is input into the new preset deep neural network model to obtain a new outbound call strategy vector;

[0031] A new outbound call task list is generated based on the preset dynamic control rules and the new outbound call strategy vector, and outbound calls are made according to the new outbound call task list.

[0032] Furthermore, to achieve the above objectives, this application also proposes a predictive outbound calling device, the device comprising:

[0033] The feature extraction module is used to perform multi-dimensional feature extraction on the collected outbound call business data in the time dimension, spatial dimension and behavioral dimension to obtain the target features of the outbound call business data in each dimension.

[0034] The weighted fusion module is used to perform weighted fusion of the target features through an attention mechanism to obtain multi-dimensional target features;

[0035] The strategy generation module is used to input the target multidimensional features into a preset deep neural network model to obtain the outbound call strategy vector.

[0036] The predictive outbound call module is used to generate an outbound call task list based on preset dynamic control rules and the outbound call strategy vector, and to make outbound calls according to the outbound call task list.

[0037] In addition, to achieve the above objectives, this application also proposes a predictive outbound calling device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the predictive outbound calling method as described above.

[0038] In addition, to achieve the above objectives, this application also proposes a storage medium that is a computer-readable storage medium, on which a computer program is stored, which, when executed by a processor, implements the steps of the predictive outbound calling method described above.

[0039] In addition, to achieve the above objectives, this application also proposes a computer program product comprising a computer program that, when executed by a processor, implements the steps of the predictive outbound calling method as described above.

[0040] This application proposes a predictive outbound calling method, apparatus, device, storage medium, and computer program product. The method includes: extracting multi-dimensional features from collected outbound calling service data in time, space, and behavior dimensions to obtain target features of the outbound calling service data in each dimension; weighting and fusing the target features through an attention mechanism to obtain target multi-dimensional features; inputting the target multi-dimensional features into a preset deep neural network model to obtain an outbound calling strategy vector; generating an outbound calling task list based on preset dynamic control rules and the outbound calling strategy vector; and performing outbound calls according to the outbound calling task list. Because this application extracts multi-dimensional features from collected outbound calling service data in time, space, and behavior dimensions to obtain target features of the outbound calling service data in each dimension, then weights and fuses the target features through an attention mechanism to obtain target multi-dimensional features, inputs the target multi-dimensional features into a preset deep neural network model to obtain an outbound calling strategy vector, and finally generates an outbound calling task list based on preset dynamic control rules and the outbound calling strategy vector, and performs outbound calls according to the outbound calling task list. In this application, target features are obtained by extracting multidimensional features from outbound call business data in different dimensions, and the target features are weighted and fused through an attention mechanism. A pre-set deep neural network model is used to perform nonlinear modeling of the multidimensional features, taking into account the nonlinear correlation between the multidimensional features, thereby obtaining the outbound call strategy vector and improving the efficiency of predictive outbound calls. Attached Figure Description

[0041] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0042] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a flowchart of the first embodiment of the predictive outbound calling method proposed in this application;

[0044] Figure 2 This is a flowchart of a second embodiment of the predictive outbound calling method proposed in this application;

[0045] Figure 3 This is a flowchart of the third embodiment of the predictive outbound calling method proposed in this application;

[0046] Figure 4 A diagram of a predictive outbound calling device provided in an embodiment of this application;

[0047] Figure 5 This is a schematic diagram of a predictive outbound calling device suitable for implementing embodiments of this application.

[0048] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0049] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of this application and are not intended to limit this application.

[0050] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0051] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0052] Understandably, predictive outbound calling is a call center automation technology based on intelligent algorithms and real-time data. Its core objective is to maximize agent utilization, improve outbound calling efficiency, and reduce operating costs while ensuring customer experience.

[0053] Existing technologies generally use linear statistical models to obtain outbound call strategy vectors, including: number of outbound calls, number priority, agent allocation ratio, predicted connection rate, and agent utilization rate. The outbound call task list is determined by the above outbound call strategy vectors, and outbound calls are made according to the outbound call task list. However, since linear statistical models can only support obtaining the correlation between low-dimensional features to obtain outbound call strategy vectors, the proportion of invalid outbound calls is high, resulting in low efficiency of predictive outbound calls.

[0054] Therefore, to address the technical challenge of improving the efficiency of predictive outbound calling, this embodiment proposes a predictive outbound calling method. The method includes: extracting multi-dimensional features from collected outbound calling business data across time, space, and behavior dimensions to obtain target features for each dimension; weighting and fusing these target features using an attention mechanism to obtain multi-dimensional target features; inputting these multi-dimensional target features into a preset deep neural network model to obtain an outbound calling strategy vector; generating an outbound calling task list based on preset dynamic control rules and the outbound calling strategy vector; and performing outbound calls according to the task list. In this embodiment, when performing predictive outbound calling, multi-dimensional features are extracted from collected outbound calling business data across time, space, and behavior dimensions to obtain target features for each dimension. Subsequently, these target features are weighted and fused using an attention mechanism to obtain multi-dimensional target features. These multi-dimensional target features are then input into a preset deep neural network model to obtain an outbound calling strategy vector. Finally, an outbound calling task list is generated based on preset dynamic control rules and the outbound calling strategy vector, and outbound calls are performed according to the task list. In this embodiment, target features are obtained by extracting multi-dimensional features from outbound call business data in different dimensions. The target features are then weighted and fused using an attention mechanism. A pre-set deep neural network model is used to perform non-linear modeling of the multi-dimensional features, taking into account the non-linear correlation between the multi-dimensional features, thereby obtaining the outbound call strategy vector and improving the efficiency of predictive outbound calls.

[0055] For ease of understanding, the following is combined with Figures 1 to 5 The predictive outbound calling method provided in the embodiments of this application, as well as the predictive outbound calling method, apparatus, device, storage medium, and computer program product provided in the following embodiments, will be described in detail.

[0056] This application provides a predictive outbound calling method, referring to... Figure 1 , Figure 1 This is a flowchart of the first embodiment of the predictive outbound calling method proposed in this application.

[0057] like Figure 1 As shown, the method includes:

[0058] Step S10: Based on the collected outbound call service data, perform multi-dimensional feature extraction in the time dimension, spatial dimension and behavioral dimension to obtain the target features of the outbound call service data in each dimension.

[0059] It should be noted that the executing entity in this embodiment can be a multifunctional machine device with predictive outbound calling capabilities, such as a predictive outbound calling device, or a device capable of performing the aforementioned functions. This embodiment uses a predictive outbound calling device (hereinafter referred to as the device) for illustration.

[0060] It should also be noted that the aforementioned multidimensional feature extraction can be a process of feature analysis of outbound call business data from multiple dimensions. The outbound call business data can be various data generated during the telephone outbound call process, such as real-time collection of agent status (idle / on / post-call processing), average call duration, answering speed, number connection rate, historical call records (reasons for success / failure), user profiles (age / region / behavioral tags), current network latency, time period characteristics (such as weekday / weekend), and promotional activity tags. The aforementioned time dimension can be a dimension that arranges and analyzes outbound call business data in chronological order, focusing on the characteristics and trends of outbound call activities in different time periods, such as different times of day or the time interval between consecutive outbound calls, to understand the impact of time factors on outbound call effectiveness.

[0061] The aforementioned spatial dimension can analyze outbound call data from the perspective of geographical region or spatial distribution, considering the impact of factors such as users or agents in different regions and locations on outbound call services, such as the connection rate of users in different cities and the working conditions of agents in different locations. The aforementioned behavioral dimension can focus on the behavioral performance of users and agents in outbound call services, analyzing user call-answering habits, agent operating methods, and work behaviors, such as user call answering frequency, reasons for call rejection, agent dialing speed, and communication skills, to understand the role of behavioral factors in outbound call services. The aforementioned target features can be features obtained through a multi-dimensional feature extraction process under a specific analytical purpose, effectively reflecting the key characteristics and influencing factors of outbound call services, such as time period coding, outbound call time intervals, agent grouping (skill group ID), number location coding, user historical connection count, and agent-user matching degree (such as language preference).

[0062] In its implementation, the aforementioned equipment analyzes outbound call data from multiple dimensions to extract features. In the time dimension, it analyzes outbound calls and connection rates for different time periods; in the spatial dimension, it examines user characteristics and agent distribution in different regions; and in the behavioral dimension, it studies user answering habits and agent operating methods. For example, in the time dimension, outbound call connection rates are higher between 9-11 AM and user call durations are longer between 8-10 PM; in the spatial dimension, business numbers have higher connection rates in first-tier cities, and agents are grouped by skill level; in the behavioral dimension, successfully connected users often have a history of business transactions, and agents spend a longer average time answering calls from high-value users. Based on this, the equipment extracts target features across each dimension, providing a precise basis for outbound call strategy formulation.

[0063] Furthermore, in order to achieve accurate predictive outbound calling, the step of extracting multi-dimensional features from the collected outbound calling service data in the time, space, and behavioral dimensions to obtain the target features of the outbound calling service data in each dimension includes:

[0064] Step S11: Extract features from the collected outbound call service data in the time dimension to obtain at least the time period code and outbound call time interval of the outbound call service data in the time dimension.

[0065] It should be noted that the above time period coding can be based on dividing a day into different time periods, assigning a unique code to each time period to represent and distinguish different time periods. For example, a day can be divided into three time periods: morning, afternoon, and evening, coded as 0, 1, and 2 respectively. The above outbound call interval can be the time difference between two consecutive outbound calls, reflecting the frequency and interval of outbound calls. For example, if a user is called twice consecutively in one day, at 9:00 and 10:00 respectively, then the outbound call interval is 1 hour.

[0066] In its implementation, the device analyzes the collected outbound call data, extracts the coding features of different time periods, divides the 24 hours of a day into multiple time periods, such as the morning (8:00-12:00), afternoon (14:00-18:00), evening (19:00-22:00) on weekdays, and the entire day on weekends, and assigns a unique coding identifier to each time period, thereby accurately distinguishing the outbound call features of different time periods.

[0067] Meanwhile, the device also calculates the outbound call time interval characteristics, calculates the difference between the timestamps of consecutive outbound call records, and obtains the time interval between adjacent outbound calls. For example, if two outbound calls to a user group are made at 9:00 and 10:00 respectively, then the corresponding outbound call time interval is 1 hour.

[0068] Step S12: Extract features from the outbound call service data in the spatial dimension to obtain at least the agent grouping and number location code of the outbound call service data in the spatial dimension.

[0069] It should be noted that the above-mentioned agent grouping can be based on factors such as agent skills, business type, or work area, dividing agents into different groups. The above-mentioned number location code can be a specific coding format for the user's geographical area.

[0070] In its implementation, the aforementioned device first groups the agents based on factors such as their skills, business type, or work area, categorizing them into different groups, such as sales, customer service, and technical support. Simultaneously, the device encodes the geographical location of user numbers, determining the user's geographic area based on the first few digits of the phone number or the region field in the user's registration information, and converting it into a specific encoding format; for example, Beijing is encoded as 010, and Shanghai as 021.

[0071] Step S13: Extract features from the outbound call service data in the behavioral dimension to obtain at least the user's historical call connection count and agent-user matching degree in the behavioral dimension of the outbound call service data.

[0072] It should be noted that the aforementioned "user's historical call count" refers to the number of times a user has successfully answered calls in past outbound calling activities, and the "agent-user matching degree" refers to the degree of matching between the agent and the user. In specific implementation, the device will count the user's historical call count and analyze the user's answering habits. Simultaneously, based on user preferences and agent skills, it will calculate the agent-user matching degree to ensure efficient communication. For example, if a user frequently answers outbound calls for financial advice, the device will count their call count and mark them as a high-intent customer. When initiating another financial-related outbound call, it will prioritize assigning an agent skilled in financial communication, improving service accuracy and user satisfaction.

[0073] Step S20: Weighted fusion of the target features is performed using an attention mechanism to obtain multidimensional target features.

[0074] It should be noted that the aforementioned attention mechanism can be a mechanism that simulates human visual attention. Through learning, it automatically assigns different weights to different features, enabling the model to focus on features more important to the current task, thereby improving model performance and generalization ability. The weighted fusion mentioned above can be a process of assigning a weight to each feature based on its importance or relevance, and then combining multiple features through weighted calculation. The aforementioned target multidimensional features can refer to the feature set obtained after weighted fusion processing, which comprehensively reflects the key characteristics of things in multiple dimensions.

[0075] In its implementation, the aforementioned device uses a pre-defined weighted algorithm model to perform weighted fusion processing on each target feature, thereby obtaining multi-dimensional target features. Taking user A as an example, their outbound call connection rate is higher between 9-11 AM, they live in a first-tier city and have a habit of answering calls, and their corresponding feature weights are: answering habit weight 0.6, regional feature weight 0.3, and time period feature weight 0.1. The device first normalizes these target features, and then calculates the weighted multi-dimensional target feature value based on the weights, i.e., 0.6×0.9+0.3×0.8+0.1×0.7=0.85, thus representing user A's comprehensive outbound call characteristics.

[0076] Step S30: Input the target multidimensional features into a preset deep neural network model to obtain the outbound call strategy vector.

[0077] It should be noted that the aforementioned preset deep neural network model can be a pre-designed and trained deep neural network model used to analyze and process the multi-dimensional features of the input. The aforementioned outbound call strategy vector can be output by the deep neural network model, containing a series of outbound call strategy vectors to guide outbound call behavior, such as the number of outbound calls, number priority, agent allocation, and other information.

[0078] In its implementation, the device integrates multi-dimensional features into a vector input to a deep neural network model, which then calculates and outputs an outbound calling strategy vector. This vector contains information such as the number of outbound calls, number priority, and agent allocation. For example, the device integrates multi-dimensional features such as user characteristics and agent status during a certain period into an input vector. The outbound calling strategy vector output by the model after calculation suggests: increasing the number of outbound calls during that period, prioritizing calls to high-value numbers, and allocating agents reasonably.

[0079] Furthermore, it should be noted that the aforementioned preset deep neural network model can be obtained by training the initial deep neural network model using historical outbound call strategy vectors and historical target multidimensional features. The aforementioned historical outbound call strategy vectors can be vector representations of past outbound call strategies, including information such as the number of historical outbound calls, historical number priority, and historical agent allocation. The aforementioned historical target multidimensional features can be multidimensional feature vectors extracted and integrated from past outbound call operations.

[0080] In its implementation, the device first needs to collect a large amount of historical outbound call data, including information such as the call time, recipient, and result, as well as corresponding multi-dimensional features of the target. Then, the device uses this data as a training set to input into an initial deep neural network model. The model automatically learns the patterns and relationships within the data and adjusts the network's weights and parameters. During training, the device continuously evaluates the model's performance, calculates the error between the predicted and actual results, and optimizes based on error feedback. Finally, when the model achieves a certain level of accuracy and stability on the training set, the device obtains a pre-defined deep neural network model.

[0081] Step S40: Generate an outbound call task list based on the preset dynamic control rules and the outbound call strategy vector, and make outbound calls according to the outbound call task list.

[0082] It should be noted that the aforementioned preset dynamic control rules can be pre-set rules used to adjust and control the outbound calling process. The aforementioned outbound calling task list can be a specific list of outbound calling tasks generated based on the outbound calling strategy vector and dynamic control rules, including detailed information such as the numbers to be called, the calling order, and the assigned agents.

[0083] In its implementation, the device determines specific outbound calling tasks based on suggestions in the outbound calling strategy vector, such as the number of outbound calls and number priority, while also referencing dynamic control rules, such as thresholds for agent utilization and connection rate. For example, if the outbound calling strategy vector suggests 100 outbound calls, prioritizing number A, and the dynamic control rules stipulate that agent utilization should not exceed 90%, the device will generate an outbound calling task list starting with the highest priority number, while ensuring that agent utilization complies with the rules. Subsequently, the device initiates outbound calls sequentially according to this list. If a number cannot be connected, it skips that number and continues to the next number according to the rules.

[0084] In this embodiment, when performing predictive outbound calls, multi-dimensional feature extraction is performed on the collected outbound call service data in the time, spatial, and behavioral dimensions to obtain target features of the outbound call service data in each dimension. Then, an attention mechanism is used to weightedly fuse these target features to obtain target multi-dimensional features. These target multi-dimensional features are then input into a preset deep neural network model to obtain an outbound call strategy vector. Finally, an outbound call task list is generated based on preset dynamic control rules and the outbound call strategy vector, and outbound calls are performed according to the outbound call task list. This embodiment improves the efficiency of predictive outbound calls by extracting target features from the outbound call service data in different dimensions, weighting and fusing these target features through an attention mechanism, and using a preset deep neural network model to non-linearly model the multi-dimensional features, taking into account the non-linear correlation between multi-dimensional features.

[0085] Based on the first embodiment, in the second embodiment, the content that is the same as or similar to that in Embodiment 1 above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 2 , Figure 2 This is a flowchart of a second embodiment of the predictive outbound calling method proposed in this application. Further, to achieve more accurate predictive outbound calling, the step of weighted fusion of the target features through an attention mechanism to obtain multi-dimensional target features includes:

[0086] Step S21: Determine the time feature vector, spatial feature vector, and behavioral feature vector based on each of the target features.

[0087] It should be noted that the aforementioned time feature vector can be a vector formed by integrating target features in the time dimension according to certain rules, used to describe the time characteristics of outbound call services. The aforementioned spatial feature vector can be a vector formed by integrating target features in the spatial dimension according to certain rules, used to describe the spatial characteristics of outbound call services. The aforementioned behavioral feature vector can be a vector formed by integrating target features in the behavioral dimension according to certain rules, used to describe the behavioral characteristics of outbound call services. Furthermore, it should be noted that each of the above dimensions can include multiple target features; for example, the target features under the time feature dimension include time period encoding and outbound call time intervals.

[0088] In its implementation, the aforementioned device first extracts time-related features from the target features, such as time period codes and outbound call intervals, arranging them sequentially to form a time feature vector that intuitively presents the time characteristics of the outbound call service. Similarly, in the spatial dimension, it extracts spatial features such as agent grouping and number location codes, integrating them to form a spatial feature vector that reflects the spatial layout of the outbound call service. In the behavioral dimension, it extracts features such as user answering habits and agent answering speed, integrating them to form a behavioral feature vector that displays the dynamic behavioral patterns of the outbound call service.

[0089] Step S22: Horizontally concatenate the temporal feature vector, the spatial feature vector, and the behavioral feature vector to generate an original feature matrix, and determine the attention weight corresponding to each target feature based on the attention mechanism and the original feature matrix.

[0090] It should be noted that the aforementioned original feature matrix can be a matrix formed by horizontally concatenating temporal, spatial, and behavioral feature vectors, used to comprehensively describe the multidimensional characteristics of outbound call services. The aforementioned attention mechanism can be a mechanism that simulates human visual attention, automatically assigning different weights to different features through learning, enabling the model to focus on features more important to the current task.

[0091] In its implementation, the aforementioned device uses an attention mechanism to determine the attention weights of target features based on the importance of each feature. For example, the device extracts features from user A's outbound call data, obtaining a time feature vector T=[0, 1] (representing outbound calls at different times), a spatial feature vector S=[1, 0] (representing the outbound call distribution of different agent groups), and a behavioral feature vector B=[1, 0.8] (representing the user's answer probability and the agent's answer speed). These vectors are then concatenated horizontally to generate the original feature matrix M=[T, S, B]=[0, 1, 1, 0, 1, 0.8]. The attention mechanism is used to calculate the weights of each feature, assuming a weight vector W=[0.3, 0.2, 0.5], representing the importance of time, spatial, and behavioral features in the decision-making process.

[0092] Furthermore, in order to obtain a more accurate original feature matrix and thus improve outbound call efficiency, the step of horizontally concatenating the time feature vector, the spatial feature vector, and the behavioral feature vector to generate the original feature matrix includes:

[0093] Step S221: Perform vector length detection on the time feature vector, the spatial feature vector, and the behavioral feature vector respectively.

[0094] Vector length detection involves counting the number of elements in a vector to determine if the vector contains complete information and whether the lengths of all vectors are consistent. In practice, the device first obtains the length of each feature vector, i.e., the number of elements in the vector, to measure the amount of information contained in that feature vector. For example, a temporal feature vector with a length of 3 indicates 3 temporal features; a spatial feature vector with a length of 2 indicates 2 spatial features; and a behavioral feature vector with a length of 4 indicates 4 behavioral features. The device then determines whether the lengths of all vectors are consistent.

[0095] Step S222: When the length detection result shows that the vector lengths are inconsistent, the time feature vector, the spatial feature vector, and the behavioral feature vector are filled based on a preset fixed length and a preset padding value, respectively.

[0096] It should be noted that the aforementioned preset fixed length can be a pre-defined target length for the feature vectors, used to unify the lengths of different feature vectors. The aforementioned preset padding value can be a specified value used to pad the feature vectors; when the feature vector length is less than the preset fixed length, this value is used to pad it to the target length.

[0097] In practice, the device first determines a preset fixed length, such as 5, and then sets a preset padding value, such as 0. If the length of the time feature vector is 3, the device pads it with two zeros to make its length 5. Similarly, the same operation is performed on the spatial feature vector and the behavioral feature vector to ensure that all feature vectors have the same length, providing a unified data format for subsequent feature processing.

[0098] Step S223: The filled time feature vector, spatial feature vector and behavioral feature vector are horizontally concatenated to generate the original feature matrix.

[0099] In its implementation, the device horizontally arranges the temporal feature vector, spatial feature vector, and behavioral feature vector in sequence to form a matrix structure. For example, the padded temporal feature vector is: T′=[0, 1, 0.5, 0, 0]; the spatial feature vector is S′=[1, 0, 0, 0, 0]; and the behavioral feature vector is B′=[1, 0.8, 0.6, 0.7, 0]. The device then horizontally concatenates these three vectors to generate the original feature matrix:

[0100] ;

[0101] By using this horizontal stitching method, the device integrates feature information from different dimensions, providing a unified matrix representation for subsequent feature analysis and processing.

[0102] Step S23: Based on the attention weights, perform weighted fusion on each of the target features, and normalize the weighted fusion result to obtain multidimensional target features.

[0103] It should be noted that the attention weights mentioned above can be weights calculated through an attention mechanism, representing the importance of each feature in the decision-making process. The weighted fusion mentioned above can be a process of assigning a weight to each feature based on its importance or relevance, and then combining multiple features through weighted calculation. The normalization process mentioned above can be a process of scaling the data proportionally to make it fall into a specific interval (such as [0, 1]) to facilitate comparison and comprehensive analysis between different features.

[0104] In its implementation, the device weights and fuses target features based on attention weights, then normalizes the results to obtain multi-dimensional target features. Assuming a weight of 0.4 for temporal features, 0.3 for spatial features, and 0.3 for behavioral features, the device first weights the temporal feature vector T=[0.5, 0.3], spatial feature vector S=[0.8, 0.2], and behavioral feature vector B=[0.6, 0.4] according to their respective weights. The calculation formula is as follows:

[0105] Weighted result = (T×0.4) + (S×0.3) + (B×0.3);

[0106] Right now:

[0107] =[0.5×0.4,0.3×0.4]+[0.8×0.3,0.2×0.3]+[0.6×0.3,0.4×0.3]

[0108] =[0.2,0.12]+[0.24,0.06]+[0.18,0.12]

[0109] =[0.62,0.3];

[0110] Then the weighted result is normalized using the following formula:

[0111] ;

[0112] The calculation yields:

[0113] ;

[0114] Ultimately, the target multidimensional features [1, 0] are obtained, which characterize the key features of outbound call business and provide a basis for subsequent strategy formulation.

[0115] Based on the first and second embodiments, in the third embodiment, the content that is the same as or similar to that in Embodiments 1 and 2 above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 3 , Figure 3 This is a flowchart of the third embodiment of the predictive outbound calling method proposed in this application. Further, the step of generating an outbound task list based on preset dynamic control rules and the outbound calling strategy vector includes:

[0116] Step S41: Obtain the predicted connection rate and predicted agent utilization rate from the outbound call strategy vector.

[0117] It should be noted that the predicted connection rate mentioned above can be the proportion of the number of calls expected to be successfully connected out of the total number of outbound calls, reflecting the expected effect of the outbound calling campaign. The predicted agent utilization rate mentioned above can be the proportion of the total working time of the agents expected to spend, reflecting the efficiency of agent resource utilization.

[0118] Step S42: Determine number priority based on preset dynamic control rules and the predicted connection rate, and determine agent allocation scheme based on preset dynamic control rules and the predicted agent utilization rate.

[0119] It should be noted that the aforementioned number priority can be a priority for outbound call numbers determined based on indicators such as predicted connection rate, used to guide the order of outbound calls. The aforementioned agent allocation scheme can be an agent allocation scheme determined based on indicators such as predicted agent utilization rate, used to rationally allocate agent resources.

[0120] Step S43: Generate an outbound call task list based on the number priority and the agent allocation scheme.

[0121] It should be noted that the aforementioned outbound call task list can be a specific outbound call task list generated based on number priority and agent allocation scheme, including detailed information such as the numbers to be called, the calling order, and the assigned agents. In specific implementation, the device first obtains the predicted connection rate and predicted agent utilization rate from the outbound call strategy vector. Numbers with higher predicted connection rates have higher priority, and agent groups with higher predicted agent utilization rates will be assigned more tasks. For example, the device sets the priority of numbers with predicted connection rates greater than 0.7 as high and those less than 0.5 as low. If the predicted connection rate of number A is 0.8 and that of number B is 0.4, then number A will be called first. For agent allocation, if the predicted agent utilization rate of agent group 1 is 0.8 (above the threshold of 0.7) and that of agent group 2 is 0.6 (below the threshold of 0.7), then more outbound call tasks will be assigned to agent group 1. Finally, the device generates and executes the outbound call task list accordingly.

[0122] Furthermore, after the step of making outbound calls according to the outbound call task list, the method further includes:

[0123] Obtain outbound call results, and determine the actual connection rate, actual agent utilization rate, and actual call loss rate based on the outbound call results;

[0124] Incremental samples are determined based on preset learning parameters, the actual connection rate, the actual agent utilization rate, and the actual call loss rate.

[0125] The preset deep neural network model is updated based on the incremental samples to obtain a new preset deep neural network model.

[0126] It should be noted that the above outbound call results can be the result data obtained after the outbound call task is completed, including information such as the number of successfully connected calls, agent working hours, and the number of unconnected calls. The above actual connection rate can be the proportion of the number of successfully connected calls to the total number of outbound calls. The above actual agent utilization rate can be the proportion of the actual agent working hours to the total available time. The above actual call loss rate can be the proportion of the number of calls that were not connected due to insufficient agents to the total number of calls that should have been connected, reflecting the call loss situation during the outbound call process.

[0127] The aforementioned preset learning parameters can be pre-defined parameters used to determine incremental samples, specifying the selection criteria for incremental samples. The aforementioned incremental samples can be samples selected from outbound call result data that meet the preset learning parameter conditions, used to update the model. The aforementioned preset deep neural network model can be a pre-designed and trained deep neural network model used to analyze and process the multidimensional features of the input to achieve specific tasks or functions. The aforementioned new preset deep neural network model can be a deep neural network model updated with incremental samples, possessing better performance and adaptability.

[0128] In its implementation, the aforementioned device collects outbound call results after the task is completed, and calculates the actual connection rate, actual agent utilization rate, and actual call loss rate accordingly. The device combines preset learning parameters with actual outbound call metrics to determine incremental samples. These incremental samples are then used to update the preset deep neural network model. The incremental samples are added to the original training dataset to retrain the model, adjusting its weights and parameters to better adapt to the new data features. For example, the device may merge selected incremental samples with previous historical data to retrain the deep neural network model.

[0129] Subsequently, when generating new target multidimensional features, the new target multidimensional features are input into a new preset deep neural network model to obtain a new outbound call strategy vector; a new outbound call task list is generated based on the preset dynamic adjustment rules and the new outbound call strategy vector, and outbound calls are made according to the new outbound call task list.

[0130] This embodiment also provides a first embodiment of a predictive outbound calling device, please refer to... Figure 4 , Figure 4 This is a diagram of a predictive outbound calling device provided in an embodiment of this application. The predictive outbound calling device includes:

[0131] The feature extraction module is used to perform multi-dimensional feature extraction on the collected outbound call business data in the time dimension, spatial dimension and behavioral dimension to obtain the target features of the outbound call business data in each dimension.

[0132] The weighted fusion module is used to perform weighted fusion of the target features through an attention mechanism to obtain multi-dimensional target features;

[0133] The strategy generation module is used to input the target multidimensional features into a preset deep neural network model to obtain the outbound call strategy vector.

[0134] The predictive outbound call module is used to generate an outbound call task list based on preset dynamic control rules and the outbound call strategy vector, and to make outbound calls according to the outbound call task list.

[0135] The feature extraction module is further configured to extract features from the collected outbound call service data in the time dimension, at least obtaining the time period code and outbound call time interval of the outbound call service data in the time dimension; extract features from the outbound call service data in the spatial dimension, at least obtaining the agent grouping and number location code of the outbound call service data in the spatial dimension; and extract features from the outbound call service data in the behavioral dimension, at least obtaining the user's historical call connection count and agent-user matching degree of the outbound call service data in the behavioral dimension.

[0136] Referring to the first embodiment of the predictive outbound calling device, this embodiment also proposes a second embodiment of the predictive outbound calling device. The contents that are the same as or similar to those in the first embodiment of the predictive outbound calling device can be referred to the above description, and will not be repeated hereafter.

[0137] The weighted fusion module is further configured to determine a temporal feature vector, a spatial feature vector, and a behavioral feature vector based on each of the target features; horizontally concatenate the temporal feature vector, the spatial feature vector, and the behavioral feature vector to generate an original feature matrix; determine the attention weight corresponding to each of the target features based on the attention mechanism and the original feature matrix; perform weighted fusion on each of the target features based on the attention weights; and normalize the weighted fusion result to obtain multidimensional target features.

[0138] The weighted fusion module is further configured to perform vector length detection on the time feature vector, the spatial feature vector, and the behavioral feature vector respectively; when the length detection result shows that the vector lengths are inconsistent, the time feature vector, the spatial feature vector, and the behavioral feature vector are filled based on a preset fixed length and a preset padding value respectively; the filled time feature vector, spatial feature vector, and behavioral feature vector are then horizontally concatenated to generate the original feature matrix.

[0139] Referring to the first and second embodiments of the predictive outbound calling device, this embodiment also proposes a third embodiment of the predictive outbound calling device. The contents that are the same as or similar to those in the first and second embodiments of the predictive outbound calling device can be referred to the above description and will not be repeated hereafter.

[0140] The predictive outbound call module is further configured to obtain the predicted connection rate and predicted agent utilization rate in the outbound call strategy vector; determine the number priority based on the preset dynamic control rules and the predicted connection rate; determine the agent allocation scheme based on the preset dynamic control rules and the predicted agent utilization rate; and generate an outbound call task list according to the number priority and the agent allocation scheme.

[0141] The predictive outbound call module is further configured to acquire outbound call results, determine the actual connection rate, actual agent utilization rate, and actual call loss rate based on the outbound call results; determine incremental samples according to preset learning parameters, the actual connection rate, the actual agent utilization rate, and the actual call loss rate; update the preset deep neural network model based on the incremental samples to obtain a new preset deep neural network model; when a new target multidimensional feature is generated, input the new target multidimensional feature into the new preset deep neural network model to obtain a new outbound call strategy vector; generate a new outbound call task list based on the preset dynamic control rules and the new outbound call strategy vector, and perform outbound calls according to the new outbound call task list.

[0142] The predictive outbound calling device provided in this embodiment employs the predictive outbound calling method described in the above embodiments, and can solve the technical problem of how to improve the efficiency of predictive outbound calling. Compared with the prior art, the beneficial effects of the predictive outbound calling device provided in this embodiment are the same as those of the predictive outbound calling method described in the above embodiments, and other technical features in the predictive outbound calling device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0143] This embodiment provides a predictive outbound calling device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the predictive outbound calling method in the first embodiment described above.

[0144] The following is for reference. Figure 5 , Figure 5 This is a schematic diagram of a predictive outbound calling device suitable for implementing embodiments of this application. The predictive outbound calling device in this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5 The predictive outbound calling device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0145] like Figure 5As shown, the predictive outbound calling device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the predictive outbound calling device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. Communication device 1009 allows the predictive outbound calling device to communicate wirelessly or wiredly with other devices to exchange data. While the figure shows predictive outbound calling devices with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0146] Specifically, according to this embodiment, the process described above with reference to the flowchart can be implemented as a computer software program. For example, this embodiment includes a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the disclosed embodiments of this embodiment.

[0147] The predictive outbound calling device provided in this embodiment, employing the predictive outbound calling method described in the above embodiments, can solve the technical problem of how to improve the efficiency of predictive outbound calling. Compared with the prior art, the beneficial effects of the predictive outbound calling device provided in this embodiment are the same as those of the predictive outbound calling method described in the above embodiments, and other technical features of this predictive outbound calling device are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0148] It should be understood that the various parts disclosed in this embodiment can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0149] The above description is merely a specific implementation of this embodiment, but the protection scope of this embodiment is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this embodiment should be included within the protection scope of this embodiment. Therefore, the protection scope of this embodiment should be determined by the protection scope of the claims.

[0150] This embodiment provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the predictive outbound calling method in the above embodiment.

[0151] The computer-readable storage medium provided in this embodiment may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0152] The aforementioned computer-readable storage medium may be included in the predictive outbound calling device; or it may exist independently and not be assembled into the predictive outbound calling device.

[0153] The aforementioned computer-readable storage medium carries one or more programs that, when executed by the predictive outbound calling device, cause the predictive outbound calling device to: perform predictive outbound calls.

[0154] Computer program code for performing the operations of this embodiment can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—and conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0155] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this embodiment. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0156] The modules described in this embodiment can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0157] The readable storage medium provided in this embodiment is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-described predictive outbound calling method, and can solve the technical problem of how to improve the efficiency of predictive outbound calling. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this embodiment are the same as the beneficial effects of the predictive outbound calling method provided in the above embodiments, and will not be repeated here.

[0158] The above descriptions are only some embodiments and do not limit the patent scope of this embodiment. All equivalent structural transformations made based on the technical concept of this application and the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included within the patent protection scope of this application.

Claims

1. A predictive outbound calling method, characterized by, The method comprises: based on the collected outbound service data, multi-dimensional feature extraction is performed in the time dimension, the space dimension and the behavior dimension respectively, and target features of the outbound service data in each dimension are obtained; target multi-dimensional features are obtained by weighting and fusing each target feature through an attention mechanism; the target multi-dimensional features are input into a preset deep neural network model to obtain an outbound strategy vector; an outbound task list is generated based on a preset dynamic regulation rule and the outbound strategy vector, and outbound is performed according to the outbound task list; the step of weighting and fusing each target feature through an attention mechanism to obtain target multi-dimensional features comprises: time feature vectors, space feature vectors and behavior feature vectors are determined based on each target feature; the time feature vectors, the space feature vectors and the behavior feature vectors are transversely spliced to generate an original feature matrix, and attention weights corresponding to each target feature are determined according to an attention mechanism and the original feature matrix; each target feature is weighted and fused based on the attention weights, and the weighted and fused result is normalized to obtain target multi-dimensional features; the step of transversely splicing the time feature vectors, the space feature vectors and the behavior feature vectors to generate an original feature matrix comprises: vector length detection is performed on the time feature vectors, the space feature vectors and the behavior feature vectors respectively; when the length detection result is that the vector lengths are inconsistent, the time feature vectors, the space feature vectors and the behavior feature vectors are filled based on a preset fixed length and a preset padding value respectively; the time feature vectors, the space feature vectors and the behavior feature vectors after filling are transversely spliced to generate an original feature matrix.

2. The method of claim 1, wherein, The step of performing multi-dimensional feature extraction on the collected outbound service data in the time dimension, the space dimension and the behavior dimension respectively to obtain target features of the outbound service data in each dimension comprises: feature extraction is performed on the collected outbound service data in the time dimension to obtain at least time period encoding and outbound time interval of the outbound service data in the time dimension; feature extraction is performed on the outbound service data in the space dimension to obtain at least seat grouping and number home encoding of the outbound service data in the space dimension; feature extraction is performed on the outbound service data in the behavior dimension to obtain at least user historical connection times and seat user matching degree of the outbound service data in the behavior dimension.

3. The method of claim 1, wherein, The step of generating an outbound task list based on a preset dynamic regulation rule and the outbound strategy vector comprises: a predicted connection rate and a predicted seat utilization rate in the outbound strategy vector are obtained; a number priority is determined based on a preset dynamic regulation rule and the predicted connection rate, and a seat allocation scheme is determined based on the preset dynamic regulation rule and the predicted seat utilization rate; an outbound task list is generated according to the number priority and the seat allocation scheme.

4. The method of claim 1, wherein, The step of making an outbound call according to the outbound call task list further comprises: obtaining an outbound call result, determining an actual connection rate, an actual seat utilization rate and an actual call loss rate based on the outbound call result; determining an incremental sample according to a preset learning parameter, the actual connection rate, the actual seat utilization rate and the actual call loss rate; updating the preset deep neural network model based on the incremental sample to obtain a new preset deep neural network model; when a new target multi-dimensional feature is generated, inputting the new target multi-dimensional feature into the new preset deep neural network model to obtain a new outbound call strategy vector; generating a new outbound call task list based on the preset dynamic regulation and control rule and the new outbound call strategy vector, and making an outbound call according to the new outbound call task list.

5. A predictive outbound call device, characterized by, The device comprises: a feature extraction module configured to extract multi-dimensional features in time, space and behavior dimensions based on collected outbound service data to obtain target features of the outbound service data in each dimension; a weighted fusion module configured to perform weighted fusion on each target feature through an attention mechanism to obtain a target multi-dimensional feature; a strategy generation module configured to input the target multi-dimensional feature into a preset deep neural network model to obtain an outbound call strategy vector; a prediction outbound call module configured to generate an outbound call task list based on a preset dynamic regulation and control rule and the outbound call strategy vector, and make an outbound call according to the outbound call task list; The weighted fusion module is further configured to determine a time feature vector, a space feature vector and a behavior feature vector based on each target feature, perform horizontal concatenation on the time feature vector, the space feature vector and the behavior feature vector to generate an original feature matrix, and determine an attention weight corresponding to each target feature according to an attention mechanism and the original feature matrix; perform weighted fusion on each target feature based on the attention weight, and perform normalization processing on the weighted fusion result to obtain a target multi-dimensional feature; The weighted fusion module is further configured to perform vector length detection on the time feature vector, the space feature vector and the behavior feature vector respectively; when the length detection result is that the vector lengths are inconsistent, perform padding on the time feature vector, the space feature vector and the behavior feature vector based on a preset fixed length and a preset padding value; perform horizontal concatenation on the padded time feature vector, space feature vector and behavior feature vector to generate an original feature matrix.

6. A predictive outbound call device, comprising: The device comprises a memory, a processor and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the prediction outbound call method according to any one of claims 1 to 4.

7. A storage medium, characterized by The storage medium is a computer readable storage medium, and the storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the prediction outbound call method according to any one of claims 1 to 4.

8. A computer program product, characterised in that, The computer program product comprises a computer program which, when executed by a processor, implements the steps of the predictive outbound call method according to any one of claims 1 to 4.

Images