Artificial intelligence-based intelligent task scheduling method and system for call center

Abstract

This invention discloses an intelligent task scheduling method and system for call centers based on artificial intelligence, involving an intelligent task scheduling algorithm to solve the problems of fuzzy scheduling strategies and low decision-making efficiency in the system. It includes determining the priority scheduling strategy for a single task based on data such as the location of the task occurrence and the real-time location of the pre-matched user, the similarity between the task to be assigned and the historical tasks processed by the pre-matched user, the capabilities of the pre-matched user, and the primary evaluation coefficient of the task to be assigned. Simultaneously, by introducing a secondary evaluation coefficient as an important decision factor for the multi-task scheduling system, and iterating the algorithm according to the actual scheduling situation, it ensures that tasks are rationally allocated under the multi-task scheduling strategy, guaranteeing the efficient operation of the system.

Description

[0001] This invention relates to the field of intelligent task scheduling algorithm technology, and more specifically, to an intelligent task scheduling method and system for call centers based on artificial intelligence. Background Technology

[0002] Intelligent task scheduling algorithms are primarily used to efficiently and rationally distribute various tasks to pre-matched users, enabling the system to operate efficiently, reliably, and intelligently. Widely used in various call centers, intelligent task scheduling algorithms ensure that different types of tasks are allocated to users more rationally and efficiently, greatly increasing the probability of successful task processing.

[0003] The existing technology has the following shortcomings:

[0004] While intelligent scheduling systems can theoretically automatically allocate tasks based on demand, the large volume of task data and high real-time requirements can lead to response delays in some systems, especially under high concurrency, resulting in slow task scheduling and impacting service quality. Call center operations are complex and dynamic, and external environments (system failures, emergencies, etc.) can cause rapid changes in task requirements. Existing intelligent scheduling systems often react slowly to sudden changes, lacking sufficient flexibility and adaptability.

[0005] To address the above problems, this invention proposes a solution. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide an artificial intelligence-based intelligent task scheduling method for call centers to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] An intelligent task scheduling method for call centers based on artificial intelligence, characterized by the following steps;

[0009] S1.1: Quantify the distance measure between the task to be assigned and the pre-matched user, the similarity measure of the historical processing tasks of the task to be assigned and the pre-matched user, and the pre-matched user's ability. Rank the tasks to be assigned according to the distance measure and similarity measure as evaluation dimensions, and calculate the difference between the distance measure and similarity measure of the highest priority task under the two dimensions.

[0010] S1.2: Use a data-driven approach to establish distance metrics, similarity metrics differences, and pre-matched user capability weights, and calculate the first-level evaluation coefficients using a weighted summation method;

[0011] S1.3: Based on the relationship between the critical threshold set by the historical task data and the evaluation coefficient, determine whether the highest priority task under the distance metric or the highest priority task under the similarity metric is the scheduling task.

[0012] S2.1: Prioritize the tasks to be assigned based on distance and similarity metrics as evaluation dimensions. Determine which task will be scheduled first based on the magnitude of the first-level evaluation coefficient E of the highest priority task under the two dimensions.

[0013] S2.2: With the first task as the highest priority evaluation dimension, the system calculates the absolute difference of the correlation between all tasks other than the highest priority task and the highest priority task under this dimension, and calculates the positive index of the first-level evaluation coefficient of each task; the average of the absolute values ​​of the correlation differences and the positive index of the first-level evaluation coefficient of the specific task are summed to obtain the second-level evaluation coefficient.

[0014] S2.3: Based on historical data and expert experience, set critical thresholds for the secondary evaluation coefficients. Compare the secondary evaluation coefficients of the tasks to be assigned with the critical thresholds to determine whether to change the evaluation dimension to find the optimal task. Then, iterate the algorithm using the relationships between the primary and secondary evaluation coefficients and the thresholds to determine whether a specific task should be scheduled.

[0015] In a preferred embodiment, step S1.1 includes the following:

[0016] In S1.1, the steps for determining the distance measurement between the task to be assigned and the pre-matched user are as follows: Let the latitude and longitude of the incident location be (φ1, λ1), and the latitude and longitude of the pre-matched user be (φ2, λ2), where φ is the latitude. λ is the accuracy; φ1=latitude1×(180 / π), λ1=longitude1×(180 / π); φ2=latitude1×(180 / π), λ2=longitude2×(180 / π); Calculate the difference in latitude and longitude: Δφ=φ2-φ1; Δλ=λ2-λ1; Using Havesing's formula: a=si n2(Δφ / 2)+cos(φ1)·cos(φ2)·s in2(Δλ / 2); where: si n2 represents the square of the sine, cos(φ1) and cos(φ) are the cosine values ​​of the latitudes of the two locations; the distance L between the two locations is given by the following formula: wherein: atan2 represents the angle difference between two points; the distance L is: L=R·c, where R is the radius of the Earth; in the process of calculating the similarity, the Euclidean distance is used to measure the similarity between the task to be assigned and the historical processing tasks of the pre-matched user; the evaluation index of the pre-matched user's task processing capability is the task processing duration; the difference of the task with the highest priority under each evaluation dimension is calculated under the selected evaluation dimension condition;

[0017] taking the distance metric as the evaluation dimension, calculate the difference of the distance metric factor between the tasks with the highest priority under different dimensions ΔX1=|d a1 -d b1 |, d a1 is the distance metric between the task with the highest distance priority and the pre-matched user, d b1 is the distance metric between the task with the highest similarity priority and the pre-matched user; under the evaluation dimension of similarity, the difference of the similarity metric factor between the tasks with the highest priority ΔX2=|x a1 -x b1 |; the capability metric factor of the pre-matched user is ΔX3.

[0018] In a preferred embodiment, step S1.2 includes the following contents:

[0019] performing data fitting, feature selection, regression analysis and other steps on the data in S1.1 to determine the weight of each evaluation dimension; the data set should include: the target variable, that is, the primary evaluation coefficient, and the independent variable, that is, the difference value of each evaluation dimension; select a polynomial regression model, use linear regression to fit polynomial features, and use the least square method to find the highest priority weight;

[0020] In a preferred embodiment, step S1.3 includes the following contents:

[0021] After the training is completed, extract the regression coefficients from the regression model, and use the obtained regression coefficients to calculate the evaluation coefficients; set the critical threshold E0 of the primary evaluation coefficient and calculate the primary evaluation coefficient E; when E>E0, the distance-first scheduling strategy should be selected; when E<E0, the similarity-first strategy should be selected.

[0022] In a preferred embodiment, step S2.1 includes the following contents:

[0023] In the multi-task scheduling system, the tasks to be assigned are prioritized according to the two dimensions of distance and similarity; the tasks to be assigned are prioritized according to the distance metric and the similarity metric, and the priority queues are a1, a2, a3, a4, ..., a n and b1, b2, b3, b4, ..., b n ; calculate the primary evaluation coefficient E of the event with the highest priority under the distance dimension a1And the first-level evaluation coefficient E of the highest priority event under the similarity dimension. b1 E a1 E b1 The priority scheduling scheme is determined by comparing it with the first-level evaluation coefficient.

[0024] In a preferred embodiment, step S2.2 includes the following:

[0025] Based on the scheduling scheme in S2.1, the dimension with the highest priority of the first task is determined as the evaluation dimension, and a subset of tasks to be assigned under the evaluation dimension is determined. Using the distance metric as the evaluation dimension, the absolute value of the correlation difference between each task to be assigned and the highest priority task is calculated. The absolute value of the correlation difference between tasks a2 and a1 is d. 21 =|d a2 -d a1 The absolute value of the correlation difference between tasks a3 and a1 to be assigned is d. 31 =|d a3 -d a1 | and calculate the tasks to be assigned: a4, a5, ..., a n The absolute value of the correlation difference; the positive index of the primary evaluation coefficient of the task to be assigned, with distance as the evaluation dimension, is the ratio of the primary evaluation coefficient of the task to the critical threshold of the primary evaluation coefficient; the positive index of the primary evaluation coefficient of the task to be assigned, a1, is E. a1 / E0; The positive index of the primary evaluation coefficient for task a2 to be assigned is E a2 / E0, and calculate the tasks to be assigned: a3, a4, a5, ..., a n The positive bias index. Tasks to be assigned: a1, a2, a3, ..., a n The average of the absolute values ​​of the correlation differences Secondary evaluation coefficient of task a1 to be assigned And calculate the tasks to be assigned: a2, a3, ..., a n The secondary evaluation coefficient is calculated; using similarity as the evaluation dimension, the positive bias index of the primary evaluation coefficient of the task to be assigned is the ratio of the primary evaluation coefficient threshold to the primary evaluation coefficient of each task; the absolute value of the correlation difference between each task to be assigned and the highest priority task is calculated, and the absolute value of the correlation difference between tasks b2 and b1 is r. 21 =|r b2 -r b1 The absolute value of the correlation difference between task b3 and task b1 is r. 31 =|r b3 -r b1 | and calculate r 41 r 51 ... r n1The positive bias exponent of the primary evaluation coefficient for task b1 to be assigned is E0 / E0. b1 The positive bias exponent of the primary evaluation coefficient for task b2 to be assigned is E0 / E0. b2 And calculate the tasks to be assigned b3, b4, b5, ..., b n Calculation of the positive bias index; tasks to be assigned b3, b4, b5, ..., b n The average of the absolute values ​​of the correlation differences Secondary evaluation coefficient of task a1 to be assigned And calculate the tasks to be assigned b2, b3, b4, b5, ..., b n The secondary evaluation coefficient.

[0026] In a preferred embodiment, step 2.3 includes the following:

[0027] The critical threshold D0 of the secondary evaluation coefficient is set according to the characteristics of the specific task and expert experience. If the secondary evaluation coefficient of the task to be assigned is greater than D0, the system should reselect the location metric as the evaluation dimension for scheduling. If the secondary evaluation coefficient of the task to be assigned is less than D0, the system should select the similarity metric as the evaluation dimension for scheduling. The system also determines which task to schedule first based on the relationship between the primary evaluation coefficients of the specific tasks.

[0028] An AI-based intelligent task scheduling system for call centers includes: an information acquisition module, a comprehensive analysis module, and an intelligent scheduling module.

[0029] Information acquisition module: The task address is obtained by using devices such as vehicle cameras, sensors, alarm phones or drones, and the user address is obtained by installing on the pre-matched user vehicle or equipped mobile device.

[0030] The comprehensive analysis module quantifies the distance between the task to be assigned and the pre-matched user, the similarity between the task to be assigned and the pre-matched user's historical processing tasks, and the pre-matched user's capabilities. It ranks the tasks to be assigned based on the distance and similarity measures as evaluation dimensions and calculates the difference between the distance and similarity measures of the highest priority task under the two dimensions.

[0031] Intelligent scheduling module: Based on the critical threshold of the secondary evaluation coefficient, it compares the secondary evaluation coefficient of the task to be assigned with the critical threshold to determine whether to change the evaluation dimension to find the best task; and through the relationship between the primary evaluation coefficient and the threshold and the relationship between the secondary evaluation coefficient and the threshold, it iterates the algorithm to determine whether a specific task to be assigned should be scheduled.

[0032] The technical effects and advantages of this invention, a call center intelligent task scheduling method based on artificial intelligence, are as follows: This invention uses the Havesing algorithm to determine the distance metric between the task to be assigned and the pre-matched user. Simultaneously, based on the similarity metrics of the tasks to be assigned and the pre-matched user's historical task processing, as well as the pre-matched user's capability metrics, it optimizes task matching efficiency. This solves the problem of whether the system should use a single-task scheduling strategy (selecting one task at a time) or a multi-task scheduling strategy (selecting multiple tasks at a time) in different scheduling scenarios. The system samples the location data between the task to be assigned and the pre-matched user, the similarity data between the pre-matched user's historically processed tasks and the task to be assigned, and the pre-matched user's capability data. The sampled data is processed, and the processed data is weighted and summed using a metric factor method to calculate the evaluation coefficient. This coefficient is then compared with a critical threshold to determine the priority allocation strategy for tasks. Furthermore, the weights of the evaluation coefficient calculation factors should be adjusted according to different situations, enabling the entire system to achieve efficient and balanced task scheduling and resource utilization in environments with a large number of tasks, maximizing the overall performance and response speed of the system. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the structure of an AI-based intelligent task scheduling method for call centers according to the present invention.

[0034] Figure 2 This is a schematic diagram of the structure of an AI-based intelligent task scheduling method for call centers according to the present invention.

[0035] Figure 3 This is a schematic diagram of the structure of an intelligent task scheduling system for call centers based on artificial intelligence, according to the present invention. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0037] This invention uses the Havesing algorithm to determine the distance metric between the task to be assigned and the pre-matched user. Simultaneously, it optimizes task matching efficiency based on the similarity metrics of the tasks to be assigned and the pre-matched user's historical task processing, as well as the pre-matched user's capability metrics. This addresses the issue of whether the system should employ a single-task scheduling strategy (selecting one task at a time) or a multi-task scheduling strategy (selecting multiple tasks at once) in different scheduling scenarios. The system samples location data between the task to be assigned and the pre-matched user, similarity data between the pre-matched user's historically processed tasks and the task to be assigned, and pre-matched user capability data. The sampled data is processed, and the processed data is weighted and summed using a metric factor method to calculate evaluation coefficients. These coefficients are then compared with critical thresholds to determine the priority allocation strategy for tasks. The weights of the evaluation coefficient calculation factors should be adjusted according to different situations, enabling the entire system to achieve efficient and balanced task scheduling and resource utilization in environments with a large number of tasks, maximizing the overall system performance and response speed.

[0038] Example 1

[0039] Figure 1 A flowchart of an artificial intelligence-based intelligent task scheduling method for call centers according to the present invention is provided; including:

[0040] S1.1: Quantify the distance measure between the task to be assigned and the pre-matched user, the similarity measure of the historical processing tasks of the task to be assigned and the pre-matched user, and the pre-matched user's capabilities. Rank the tasks to be assigned based on the distance measure and similarity measure as evaluation dimensions, and calculate the difference between the distance measure and similarity measure of the highest priority task under the two dimensions.

[0041] S1.2: Use a data-driven approach to establish distance metrics, similarity metrics differences, and pre-matched user capability weights, and calculate the first-level evaluation coefficients using a weighted summation method.

[0042] S1.3: Based on the relationship between the critical threshold set by the historical task data and the first-level evaluation coefficient, determine whether the highest priority task under the distance metric or the highest priority task under the similarity metric is the scheduling task.

[0043] Specifically as follows:

[0044] S1.1: To accurately obtain the distance measurement between the task to be assigned and the pre-matched user, when the task to be assigned occurs, the system uses the latitude and longitude information already included in the device report (e.g., through vehicle cameras, sensors, emergency phones, or drones) to convert place names or street addresses into latitude and longitude using the address resolution service (Geodicing) to accurately obtain the latitude and longitude of the task location. It can also obtain GPS data from the vehicle or mobile device (such as a smartphone or in-vehicle device) installed on the pre-matched user to determine the location of the pre-matched user in real time (GPS positioning data typically includes longitude, latitude, and a timestamp). Specifically, the steps to determine the distance measurement between the task to be assigned and the pre-matched user are as follows: assuming the latitude and longitude of the task to be assigned is (φ1, λ1) and the latitude and longitude of the pre-matched user is (φ2, λ2), where φ is latitude and λ is precision; the Havesing formula requires the use of radians, so the latitude and longitude need to be converted to radians. Specifically: φ1=latitude1×(180 / π), λ1=longitude1×(180 / π); φ2=latitude1×(180 / π), λ2=longitude2×(180 / π); Calculate the difference in latitude and longitude: Δφ=φ2-φ1; Δλ=λ2-λ1; Use the Havesing formula: a=sin2(Δφ / 2)+cos(φ1)·cos(φ2)·sin2(Δλ / 2); where: sin2 represents the square of the sine, and cos(φ1) and cos(φ2) are the cosine values ​​of the latitudes of the two locations. The distance L between the two locations is given by the following formula: Where: atan2 is a function that calculates the arctangent, returning the angle, representing the angular difference between two points; the distance L is: L = R·c, where R is the radius of the Earth, usually taken as 6731 kilometers. This process determines the distance between the task to be assigned and the pre-matched user based on their satellite positioning and real-time location, exhibiting general characteristics and accurately reflecting the distance measurement between the task to be assigned and the pre-matched user.

[0045] By using the similarity between pre-matched user's historically processed task types and the tasks to be assigned as an evaluation dimension for system scheduling, the system's decision-making becomes more efficient. Higher similarity indicates a higher probability that the corresponding task will be successfully processed. Specifically, in calculating similarity, Euclidean distance is used to measure the similarity between the tasks to be assigned and the pre-matched user's historically processed tasks. Specifically, multiple features of the tasks to be assigned (such as task type, task location, task time, weather conditions, road conditions, and task severity) are used as numerical features to construct the feature vector of the tasks to be assigned, and Euclidean distance is used to measure the similarity between different tasks to be assigned. First, the different features of each task to be assigned need to be numerated (vectorized), transforming the features of each task to be assigned into a numerical vector.

[0046] For example, suppose the tasks to be assigned have the following features: Task type: such as "collision," "scratching," etc., each task type can be assigned a numerical value, or one-hot encoding can be used to convert it into a binary feature; Task time: Time information can usually be converted into a continuous numerical value, for example, the time when the task occurs can be converted into the number of hours in a day, or a standard timestamp; Weather conditions: Different weather conditions (such as sunny, rainy, snowy, etc.) can be mapped to different numerical values. Alternatively, classification encoding can be used to represent weather conditions as a numerical value; Road conditions: Road conditions can be represented by multiple different indicators, such as road slipperiness, traffic flow, etc., each of which can be converted into a numerical feature; Task severity: Generally, the severity of the task can be converted into a numerical value, for example, minor = 1, severe = 2, fatal = 3. Assume the feature vector of each task to be assigned is X = (x1, x2, x3, ..., x...). n ), where x1, x2, ..., x n These are the numerical features of the task. The calculation of Euclidean distance depends on the differences between these features. Therefore, if the numerical ranges of different features differ too much (for example, the difference between road condition information and task severity), some features may have an excessive impact on the distance calculation, thus biasing the results towards the differences of certain features.

[0047] Specifically, the features need to be standardized or normalized. A common standardization algorithm is Z-Score standardization, which transforms the data into a distribution with a mean of 0 and a standard deviation of 1, ensuring that the numerical ranges of each feature are consistent. For each feature x... i Its standardization formula is: x i ′=(x i -μ i ) / σ i , where: x iIt is the i-th feature value in the original data, μ i It is the mean of the i-th feature, σ i This is the standard deviation of the i-th feature. After standardization, the values ​​of different features will be compared on the same scale, thus avoiding the dominance of certain features with large values ​​in the calculation process. Next, missing values ​​need to be handled. Missing values ​​are a common problem in real-world datasets. If they are not handled, it may lead to errors in calculating Euclidean distance or cause the calculation results to deviate from reality. To optimize the handling of missing values, the mean imputation method is used. For numerical features, the mean of the feature can be used to impute missing values, as shown in the following formula: in, It is feature x i The sum of all known values, where n is the number of known values; then calculate the Euclidean distance: Euclidean distance is a common way to measure the difference between two vectors (i.e., two tasks to be assigned). Given the feature vectors X1 = (x... 1,1 ,x 1,2 ,…,x 1，n ) and X2 = (x 2,1 ,x 2，2 ,…,x 2，n The Euclidean distance d between them is calculated using the following formula: Where X 1, and X 2,i Let represent the values ​​of the two tasks to be assigned on the i-th feature, and n be the dimension of the feature (i.e., the number of features). The smaller the Euclidean distance, the smaller the difference between the two tasks on each feature, and the higher their similarity. Conversely, the larger the Euclidean distance, the greater the difference between the two tasks, and the lower their similarity.

[0048] In the entire system, the pre-matched user's ability has a slightly positive impact on distance and similarity metrics. Experienced pre-matched users are better able to handle unexpected tasks and complex events, thus the influence of similarity metrics as an evaluation dimension for target task scheduling should be weakened. A quantitative evaluation scheme for the pre-matched user's task processing ability can use task processing time as the evaluation metric. Specifically, the time when a pre-matched user arrives at the task processing site is t1, and the task completion event is t2, i.e., the task processing time Δx3 = t2 - t1.

[0049] Specifically, for the tasks to be assigned and the pre-matched users, we can obtain: distance metric factor = 1 / distance metric, similarity metric factor = similarity metric, and capability metric factor = task processing time. Next, we should calculate the difference between the highest priority assigned tasks under each evaluation dimension, presented as a difference: for example, in the distance dimension, assuming there are tasks to be assigned e and f and a pre-matched user j, the difference (distance) = |distance metric factor(e,j) - distance metric factor(f,j)|; similarly, the difference under the similarity dimension can be calculated.

[0050] Furthermore, the tasks to be assigned are prioritized based on two evaluation dimensions: distance and similarity. The priority queues are a1, a2, a3, a4, ..., a n and b1, b2, b3, b4, ..., b n Using distance as the evaluation dimension, the difference in distance metric factors between the highest priority tasks under different dimensions is calculated, ΔX1=|d a1 -d b1 |. Where d a1 d is a distance metric between the highest priority task and the pre-matched user. b1 This is a distance measure between the highest-priority similarity task and the pre-matched user; similarly, the difference in similarity metric between the highest-priority tasks, evaluated by similarity, can be calculated as ΔX² = |x a1 -x b1 |;The capability metric factor of the pre-matched user is quantified and defined as ΔX3.

[0051] S1.2: The primary evaluation coefficient serves as the standard for determining the final scheduling strategy. The larger the absolute value of the primary evaluation coefficient, the higher the priority of the corresponding task. The standard is determined by three influencing factors. Specifically, the data in step S1.1 is subjected to data fitting, feature selection, regression analysis, and other steps to determine the weight of each evaluation dimension (the distance measure between the task to be assigned and the pre-matched user, the similarity measure of the historical tasks processed by the task to be assigned and the pre-matched user, and the processing capacity of the pre-matched user).

[0052] Specifically, data fitting first requires ensuring sufficient training data. The dataset should include: the target variable, i.e., the first-level evaluation coefficient E; and the independent variables, i.e., the differences between each evaluation dimension (e.g., ΔX1: the difference between the distance metric factor between the task to be assigned and the pre-matched user; ΔX2: the difference between the similarity metric factor between the task to be assigned and the pre-matched user's historical processing tasks; ΔX3: the pre-matched user's ability metric factor). A multinomial regression model is chosen: the basic form of the multinomial regression model is: E = w0 + w1·ΔX1 + w2·ΔX2 + w3·(ΔX1)² + w4·(ΔX2)² + w5·ΔX1·ΔX2 + ... The goal here is to fit a multinomial regression model containing first-order terms, second-order terms, and interaction terms. To construct the multinomial regression model, multinomial features need to be generated first, including: first-order terms (original features): ΔX1, ΔX2; second-order terms (squared terms): (ΔX1). 2 (ΔX2) 2 The cross term is ΔX1·ΔX2. By combining these features, a multinomial feature matrix is ​​obtained. For example, if there are three task sample data: Task Sample 1: ΔX1=2, ΔX2=3, ΔX3=1; Task Sample 2: ΔX1=5, ΔX2=6, ΔX3=2; Task Sample 3: ΔX1=8, ΔX2=4, ΔX3=3, then through feature engineering, the samples of each task are converted into the following multinomial features: Task Sample 1 features: 2,3,1,4,9,1,6,2,3; Task Sample 2 features: 5,6,2,25,36,4,30,10,12; Task Sample 3 features: 8,4,3,64,16,9,32,24,12. Among them, the first 3 columns are the original features (ΔX1, ΔX2, ΔX3), and the following columns are the second-order terms and cross features, respectively.

[0053] To fit a multinomial feature using linear regression, the least squares method can be used to find the highest priority weights w0, w1, w2, w3. The specific steps are: Constructing the multinomial feature matrix X: Convert the data for each task sample into a multinomial feature matrix X, where each row represents a task sample and includes first-order terms, second-order terms, and interaction terms. Training the regression model: Use the multinomial feature matrix X and the target variable E to train a linear regression model, obtaining the regression coefficients (weights) w0, w1, w2, w3… For multinomial regression, the highest priority regression coefficient W can be calculated using the following formula (least squares method): W = (X…) T ·X) -1 ·X T ·y, where: X is the feature matrix (including first-order, second-order and interaction terms), y is the target variable (evaluation coefficient E), and W is the regression coefficient vector.

[0054] After the training is completed, the effect of the regression model is evaluated, and the following indicators are commonly used. Mean Squared Error (MSE): used to measure the error between the predicted value and the actual value, coefficient of determination R 2 : represents the ability of the model to explain the variance of the data, and a value closer to 1 indicates a better model fitting effect. If the fitting effect of the model is good, that is, the error is small, and R 2 is relatively high, it indicates that the selected polynomial regression model can fit the data well. The regression coefficients w1, w 2… are extracted from the regression model. These coefficients represent the influence weight of each evaluation dimension on the final first-level evaluation coefficient E. In particular, assuming that w1 represents the influence weight of ΔX1 on the evaluation coefficient, w2 represents the influence weight of ΔX2 on the evaluation coefficient (which is a negative value here), and w3 represents the influence weight of ΔX3 on the evaluation coefficient, if the coefficients w4, w5... of the second-order terms or interaction terms are relatively large, it indicates that these terms also have a relatively large influence on the evaluation coefficient. Finally, the obtained regression coefficients are used to calculate the evaluation coefficient.

[0055] S1.3: First, it is necessary to determine the critical threshold of the first-level evaluation coefficient, and the critical threshold can be determined based on the expert experience method. Experts can dynamically set an appropriate critical threshold according to factors such as the distance between the task to be assigned and the pre-matched user, the similarity between the task to be assigned and the historical processed tasks of the pre-matched user, and the capability of the pre-matched user. Experts can set an empirical threshold and fine-tune it according to the actual situation; specifically, the first-level evaluation coefficient E = w1·ΔX1 + w2·ΔX2 + w3·ΔX3. It should be noted that w1 and w3 take positive values, and w2 takes a negative value, that is, E is directly proportional to ΔX1 and ΔX3, and inversely proportional to ΔX2; further, a larger value of the first-level evaluation coefficient E indicates that the distance metric priority is more suitable as the direction of the scheduling strategy, and a smaller value of the first-level evaluation coefficient E indicates that the similarity metric has greater advantages as an evaluation dimension. Assume that E0 is the critical threshold for the first-level evaluation. When E>E0, the distance metric priority scheduling strategy should be selected, that is, the task with the highest distance metric priority is the target scheduling task; when E<E0, the similarity metric priority strategy should be selected, that is, the task with the highest similarity metric priority is the target scheduling task.

[0056] Embodiment 2

[0057] In Embodiment 1 of this invention, a key example is given illustrating how to determine a single-task scheduling strategy using a weighted method. This involves determining the weight of each evaluation dimension (distance between the task to be assigned and the pre-matched user, similarity between the task to be assigned and the pre-matched user's historical task processing, and the pre-matched user's processing ability) through data fitting, feature selection, and regression analysis. The scheduling strategy is then determined by comparing the first-level evaluation coefficient with its critical threshold. This approach can quickly identify the tasks that should be scheduled in a single-task scheduling system. However, if multiple tasks need to be scheduled simultaneously, and the task sample size is large, the scheduling strategy becomes relatively ambiguous, affecting the system's decision-making efficiency. With the pre-matched user's ability as a positive factor, the priority of distance and similarity metrics as evaluation dimensions needs to be dynamically adjusted based on specific circumstances. Based on the above, Embodiment 2 further refines the task scheduling scheme in a multi-task scheduling scenario. Figure 2 A flowchart of an artificial intelligence-based intelligent task scheduling method for call centers according to the present invention is provided; including:

[0058] S2.1: Prioritize the tasks to be assigned based on distance and similarity metrics as evaluation dimensions. Determine which task will be scheduled first based on the first-level evaluation coefficient of the highest priority task under the two dimensions.

[0059] S2.2: With the first task as the highest priority evaluation dimension, the system calculates the absolute difference of the correlation between all candidate tasks other than the highest priority task and the highest priority task under this dimension, and calculates the positive index of the first-level evaluation coefficient of each task; the system then sums the average of the absolute values ​​of the correlation differences and the positive index of the first-level evaluation coefficient of the specific task to obtain the second-level evaluation coefficient.

[0060] S2.3: Based on historical data and expert experience, set critical thresholds for the secondary evaluation coefficients. Compare the secondary evaluation coefficients of the tasks to be assigned with the critical thresholds to determine whether to change the evaluation dimension to find the optimal task. Then, iterate the algorithm using the relationships between the primary and secondary evaluation coefficients and the thresholds to determine whether a specific task should be scheduled.

[0061] S2.1: In steps S1.1 and S1.2, in the multi-task scheduling system, the tasks to be assigned are similarly prioritized based on two dimensions: distance and similarity. Under the strategy of scheduling only a single task, the evaluation coefficient of the target task is determined by a weighted summation method, and its relationship with the first-level critical threshold E0 is used to decide which task should be scheduled first. Specifically, the tasks to be assigned are prioritized based on distance and similarity, with priority queues a1, a2, a3, a4, ..., a nand b1, b2, b3, b4, ..., b n ; Calculate the first-level evaluation coefficient E for the highest priority event under the distance metric dimension. a1 And the first-level evaluation coefficient E of the highest priority event under the similarity metric dimension. b1 E a1 E b1 Compare with the primary evaluation coefficient to determine which task should be prioritized for scheduling.

[0062] S2.2: Based on the scheduling scheme determined in step S2.1, the dimension with the highest priority of the first task is determined as the evaluation dimension, and a subset of tasks to be assigned under the evaluation dimension is determined. Specifically, based on the priority queue in step S2.1, the number of subsets to be assigned should be n-1; calculate the absolute value of the correlation difference between the n-1 tasks in the subset of tasks to be assigned and the highest priority task under this dimension; specifically, using distance as the evaluation dimension, calculate the absolute value of the correlation difference between each task to be assigned and the highest priority task, and the absolute value of the correlation difference between tasks a2 and a1 is d. 21 =|d a2 -d a1 The absolute value of the correlation difference between tasks a3 and a1 to be assigned is d. 31 =|d a3 -d a1 | Tasks to be assigned: a4, a5, ..., a n The absolute value of the correlation difference is calculated in a similar way to the above method, and will not be repeated here.

[0063] Furthermore, using distance as the evaluation dimension, the positive bias exponent of the primary evaluation coefficient of the task to be assigned is the ratio of the task's primary evaluation coefficient to the critical threshold of the primary evaluation coefficient. Specifically, the positive bias exponent of the primary evaluation coefficient of the task to be assigned, a1, is E. a1 / E0; The positive index of the primary evaluation coefficient for task a2 to be assigned is E a2 / E0, tasks to be assigned: a3, a4, a5, ..., a n The calculation of the positive exponent is not elaborated here. Tasks to be assigned are a1, a2, a3, ..., a... n The average of the absolute values ​​of the correlation differences Secondary evaluation coefficient of task a1 to be assigned Tasks to be assigned: a2, a3, ..., a n The calculation method for the secondary evaluation coefficient is the same as that for task a1, and will not be repeated here.

[0064] Furthermore, using similarity as the evaluation dimension, the positive bias index of the first-level evaluation coefficient of the task to be assigned is the ratio of the first-level evaluation coefficient threshold to the sum of the evaluation coefficients of each task. Specifically, based on the priority queue in step S2.1, the number of subsets to be assigned in the similarity dimension should be n-1; calculate the absolute value of the correlation difference between the n-1 tasks in the subset to be assigned and the highest priority task in this dimension; specifically, using similarity as the evaluation dimension, calculate the absolute value of the correlation difference between each task to be assigned and the highest priority task, where the absolute value of the correlation difference between tasks b2 and b1 is r. 21 =|r b2 -r b1 The absolute value of the correlation difference between task b3 and task b1 is r. 31 =|r b3 -r b1 |,r 41 r 51 ... r n1 The calculation method is the same as above, and will not be repeated here. The positive index of the primary evaluation coefficient of task b1 to be assigned is E0 / E b1 The positive bias exponent of the primary evaluation coefficient for task b2 to be assigned is E0 / E0. b2 Tasks to be assigned: b3, b4, b5, ..., b n The calculation of the positive bias index will not be elaborated here. Tasks to be assigned b3, b4, b5, ..., b n The average of the absolute values ​​of the correlation differences Secondary evaluation coefficient of task a1 to be assigned Tasks to be assigned: b2, b3, b4, b5, ..., b n The calculation method for the secondary evaluation coefficient is the same as for task b1, and will not be repeated here. It is worth noting that for tasks using distance as the evaluation dimension, the positive bias index of the secondary evaluation coefficient is directly proportional to the secondary evaluation coefficient; while for tasks using similarity as the evaluation dimension, the positive bias index of the secondary evaluation coefficient is inversely proportional to the secondary evaluation coefficient.

[0065] S2.3: The critical threshold D0 for the secondary evaluation coefficient can be set based on the characteristics of the specific task and expert experience. Specifically, when the system schedules a task to be assigned, if the secondary evaluation coefficient of the task to be assigned is greater than D0, the system should reselect location as the evaluation dimension for scheduling; if the secondary evaluation coefficient of the task to be assigned is less than D0, the system should select similarity as the evaluation dimension for scheduling. Furthermore, the priority of scheduling a task is determined based on the relationship between the primary evaluation coefficients of specific tasks.

[0066] This invention also discloses an intelligent task scheduling system for call centers based on artificial intelligence, such as... Figure 3As shown, it includes: an information acquisition module, a comprehensive analysis module, and an intelligent scheduling module;

[0067] Information acquisition module: The task address is obtained by using devices such as vehicle cameras, sensors, alarm phones or drones, and the user address is obtained by installing on the pre-matched user vehicle or equipped mobile device;

[0068] The comprehensive analysis module quantifies the distance between the task to be assigned and the pre-matched user, the similarity between the task to be assigned and the pre-matched user's historical processing tasks, and the pre-matched user's capabilities. It ranks the tasks to be assigned based on the distance and similarity measures as evaluation dimensions and calculates the difference between the distance and similarity measures of the highest priority task under the two dimensions.

[0069] Intelligent scheduling module: Based on the critical threshold of the secondary evaluation coefficient, it compares the secondary evaluation coefficient of the task to be assigned with the critical threshold of the secondary evaluation coefficient to determine whether to change the evaluation dimension to find the optimal task; and through the relationship between the primary evaluation coefficient and the threshold and the relationship between the secondary evaluation coefficient and the threshold, it iterates the algorithm to determine whether a specific task to be assigned should be scheduled.

[0070] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0071] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0072] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0073] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0074] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that cannot be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0075] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent task scheduling method for call centers based on artificial intelligence, characterized in that, comprising the following steps; S1.1: Quantify the distance metric between the task to be allocated and the pre-matched user, the similarity metric between the task to be allocated and the historical processed tasks of the pre-matched user, and the capability of the pre-matched user, take the distance metric and the similarity metric as evaluation dimensions to sort the tasks to be allocated, and calculate the difference between the distance metric and the similarity metric of the task with the highest priority under the two dimensions; S1.2: Establish the weights of the distance metric, the difference of the similarity metric and the pre-matched user capability by using a data-driven method, and calculate a first-level evaluation coefficient through a weighted summation method; S1.3: According to the relationship between the critical threshold of the first-level evaluation coefficient set based on historical task data and the first-level evaluation coefficient, determine whether the task with the highest priority under the distance metric or the task with the highest priority under the similarity metric is taken as the scheduling task; S2.1: Perform priority sorting on the tasks to be allocated by taking the distance metric and the similarity metric as evaluation dimensions, and determine which of the two tasks is to be preferentially scheduled as the first task according to the magnitude of the first-level evaluation coefficient of the task with the highest priority under the two dimensions; S2.2: Under the evaluation dimension where the first task is the task with the highest priority, calculate the absolute difference of the correlation between all remaining tasks to be allocated except the task with the highest priority and the task with the highest priority under this dimension, and calculate the positive deviation index of the first-level evaluation coefficient of each task; sum the average value of the absolute value of the correlation difference and the positive deviation index of the first-level evaluation coefficient of the specific task to obtain a second-level evaluation coefficient; S2.3: Set a critical threshold for the second-level evaluation coefficient, compare the second-level evaluation coefficient of the task to be allocated with the critical threshold of the second-level evaluation coefficient, and determine whether to replace the evaluation dimension to find an optimal task; and perform algorithm iteration through the relationship between the first-level evaluation coefficient and the critical threshold of the first-level evaluation coefficient and the relationship between the second-level evaluation coefficient and the critical threshold of the second-level evaluation coefficient to determine whether a specific task is scheduled.

2. The artificial intelligence-based intelligent task scheduling method for a call center according to Claim 1, characterized in that: In S1.1, the steps to determine the distance measurement between the task to be assigned and the pre-matched user are as follows: Let the latitude and longitude of the task be (φ1, λ1), and the latitude and longitude of the pre-matched user be (φ2, λ2), where φ is the latitude and λ is the precision; φ1 = latitude1 × (180 / π), λ1 = longitude1 × (180 / π); φ2 = latitude1 × (180 / π), λ2 = longitude2 × (180 / π); calculate the difference between the latitude and longitude: Δφ = φ2 - φ1; Δλ = λ2 - λ1; use the Havessian formula: a = sin2(Δφ / 2) + cos(φ1)·cos(φ2)·sin2(Δλ / 2); where: sin2 represents the square of the sine, cos(φ1) and cos(φ2) are the cosine values ​​of the latitudes of the two locations; the distance L between the two locations is given by the following formula: Where: atan2 represents the angle difference between two points; distance L is: L = R·c, where R is the radius of the Earth; Euclidean distance is used to measure the similarity between the task to be assigned and the historical processing tasks of the pre-matched user; the task processing ability evaluation index of the pre-matched user is the task processing time; calculate the difference of the highest priority task under the selected evaluation dimension conditions under each evaluation dimension. The distance metric is used to calculate the difference in distance metric factors between the highest priority tasks in different evaluation dimensions, ΔX1 = |d a1 -d b1 |,d a1 d is the distance metric between the highest priority task and the pre-matched user. b1 The similarity measure is the distance between the task with the highest similarity priority and the pre-matched user; similarity is the difference in similarity metrics between the highest-priority tasks in the evaluation dimension, ΔX² = |x a1 -x b1 |;The capability metric factor for pre-matched users is ΔX3.

3. The artificial intelligence-based intelligent task scheduling method for a call center according to Claim 2, characterized in that: performing steps such as data fitting, feature selection and regression analysis on the data in S1.1 to determine the weight of each evaluation dimension; the dataset shall include: a target variable which is the first-level evaluation coefficient, and independent variables; selecting a polynomial regression model; using linear regression to fit polynomial features, and using the least square method to find the weight with the highest priority; After the training is completed, extract the regression coefficient from the regression model, and use the obtained regression coefficient to calculate the first-level evaluation coefficient; set the critical threshold E0 of the first-level evaluation coefficient and calculate the first-level evaluation coefficient E; when E>E0, select the distance-first scheduling strategy; when E<E0, select the similarity-first strategy.

4. The artificial intelligence-based intelligent task scheduling method for a call center according to Claim 1, characterized in that: In a multi-task scheduling system, tasks to be assigned are prioritized based on two dimensions: distance and similarity. The tasks are then prioritized using distance and similarity metrics, with priority queues a1, a2, a3, a4, ..., a... n and b1, b2, b3, b4, ..., b n ; Calculate the first-level evaluation coefficient E for the highest priority event under the distance dimension. a1 And the first-level evaluation coefficient E of the highest priority event under the similarity dimension. b1 E a1 E b1 The priority scheduling scheme is determined by comparing it with the first-level evaluation coefficient.

5. The artificial intelligence-based intelligent task scheduling method for a call center according to Claim 4, characterized in that: Based on the scheduling scheme in S2.1, the dimension with the highest priority of the first task is determined as the evaluation dimension, and a subset of tasks to be assigned under the evaluation dimension is determined. Using the distance metric as the evaluation dimension, the absolute value of the correlation difference between each task to be assigned and the highest priority task is calculated. The absolute value of the correlation difference between tasks a2 and a1 is d. 21 =|d a2 -d a1 The absolute value of the correlation difference between tasks a3 and a1 to be assigned is d. 31 =|d a3 -d a1 | and calculate the tasks to be assigned: a4, a5, ..., a n The absolute value of the correlation difference; the positive index of the primary evaluation coefficient of the task to be assigned, with distance as the evaluation dimension, is the ratio of the primary evaluation coefficient of the task to the critical threshold of the primary evaluation coefficient; the positive index of the primary evaluation coefficient of the task to be assigned, a1, is E. a1 / E0; The positive index of the primary evaluation coefficient for task a2 to be assigned is E a2 / E0, and calculate the tasks to be assigned: a3, a4, a5, ..., a n The positive index; the tasks to be assigned a1, a2, a3, ..., a n The average of the absolute values ​​of the correlation differences Secondary evaluation coefficient of task a1 to be assigned And calculate the tasks to be assigned: a2, a3, ..., a n The secondary evaluation coefficient; using similarity measure as the evaluation dimension, the positive index of the primary evaluation coefficient of the task to be assigned is the ratio of the primary evaluation coefficient threshold to the primary evaluation coefficient of each task. Calculate the absolute value of the correlation difference between each task to be assigned and the highest priority task. The absolute value of the correlation difference between task b2 and task b1 is r. 21 =|r b2 -r b1 The absolute value of the correlation difference between task b3 and task b1 is r. 31 =|r b3 -r b1 | and calculate r 41 r 51 ... r n1 The positive bias exponent of the primary evaluation coefficient for task b1 to be assigned is E0 / E0. b1 The positive bias exponent of the primary evaluation coefficient for task b2 to be assigned is E0 / E0. b2 And calculate the tasks to be assigned b3, b4, b5, ..., b n Calculation of the positive bias index; tasks to be assigned b3, b4, b5, ..., b n The average of the absolute values ​​of the correlation differences, r = (r 21 +r 31 +r 41 +…+r -n1 ) / n; Secondary evaluation coefficient D of task a1 to be assigned a1 =E0 / E b1 +r, and calculate the tasks to be assigned b2, b3, b4, b5, ..., b n The secondary evaluation coefficient.

6. The intelligent task scheduling method for call centers based on artificial intelligence according to claim 5, characterized in that: The critical threshold D0 of the secondary evaluation coefficient is set according to the characteristics of the specific task and expert experience; if the secondary evaluation coefficient of the task to be assigned is greater than D0, the location metric is reselected as the evaluation dimension for scheduling; if the secondary evaluation coefficient of the task to be assigned is less than D0, the similarity metric is selected as the evaluation dimension for scheduling; and the priority scheduling of the task is determined according to the relationship of the primary evaluation coefficients between the specific tasks to be assigned.

7. An artificial intelligence-based intelligent task scheduling system for call centers, used to implement the artificial intelligence-based intelligent task scheduling method for call centers as described in any one of claims 1-6, characterized in that, include: Information acquisition module, comprehensive analysis module, and intelligent scheduling module; Information acquisition module: The task address is obtained by using devices such as vehicle cameras, sensors, alarm phones or drones, and the user address is obtained by installing on the pre-matched user vehicle or equipped mobile device; The comprehensive analysis module quantifies the distance between the task to be assigned and the pre-matched user, the similarity between the task to be assigned and the pre-matched user's historical processing tasks, and the pre-matched user's capabilities. It ranks the tasks to be assigned based on the distance and similarity measures as evaluation dimensions and calculates the difference between the distance and similarity measures of the highest priority task under the two dimensions. Intelligent scheduling module: Based on the critical threshold of the secondary evaluation coefficient, it compares the secondary evaluation coefficient of the task to be assigned with the critical threshold of the secondary evaluation coefficient to determine whether to change the evaluation dimension to find the optimal task; and through the relationship between the primary evaluation coefficient and the threshold and the relationship between the secondary evaluation coefficient and the threshold, it iterates the algorithm to determine whether a specific task to be assigned should be scheduled.

Images