Incoming call seat distribution method, device, equipment and medium
By screening historical call data, using content complexity prediction model and RFM model to classify customer value groups, the problem of efficient and rationality of bank customer service agent resource allocation is solved, high-quality service during peak hours is ensured, and customer experience is improved.
Patent Information
- Application Number
- CN202411862622.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-17
- Publication Date
- 2025-04-25
AI Technical Summary
How to efficiently and reasonably allocate bank customer service seat resources to ensure high-quality services are provided during peak hours.
By filtering historical customer service call data, calling content complexity prediction model and RFM model, classifying customer value groups, and reasonably allocating incoming call seats.
It has achieved efficient and reasonable allocation of agent resources during peak hours, ensured high-quality services, and improved customer experience and satisfaction.
Smart Images
Figure CN119892988A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of computer technology, and in particular to a method, device, equipment and medium for allocating incoming call seats. Background Art
[0002] By interacting directly with customers, bank customer service agents can better understand and meet customer needs, thereby improving customer experience and satisfaction.
[0003] Good customer service can enhance customers' trust in the bank and make them more willing to choose the bank for transactions and cooperation. However, given the limited seat resources, how to allocate seat resources efficiently and reasonably has become a key issue to be solved in bank operation management. Summary of the invention
[0004] The technical problem to be solved by the embodiments of the present application is to provide a method, device, equipment and medium for allocating incoming calls, so as to efficiently and reasonably allocate seat resources and ensure that high-quality services can still be provided during peak hours.
[0005] In a first aspect, an embodiment of the present application provides a method for allocating seats for incoming calls, the method comprising:
[0006] Filter the target customer service call data of the specified business scenario from the customer service call data within the historical statistical period from the current time;
[0007] Calling a pre-trained content complexity prediction model to process each customer service call data of each customer in the target customer service call data to obtain the content complexity of each customer's call each time; the content complexity is predicted by the content complexity prediction model based on the call service type and call emotion information of each customer at each call;
[0008] Determine the duration of each customer's most recent call and the number of calls within the historical statistical period based on the customer service call data of each customer in the target customer service call data;
[0009] Based on the RFM model, the duration of the most recent call, the number of calls and the average value of the content complexity of each customer in the historical statistical period are binary classified according to a preset threshold, and the value group classification results of all calling customers are obtained, so that when a customer call is received, the customer call is assigned a caller seat according to the value group classification results.
[0010] In a second aspect, an embodiment of the present application provides a device for allocating seats for incoming calls, the device comprising:
[0011] A target call data screening module is used to screen target customer service call data for a specified business scenario from customer service call data within a historical statistical period from the current time;
[0012] A content complexity acquisition module is used to call a pre-trained content complexity prediction model to process each customer service call data of each customer in the target customer service call data to obtain the content complexity of each customer's call each time; the content complexity is predicted by the content complexity prediction model based on the call service type and call emotion information of each customer at each call;
[0013] A call count determination module, used to determine the duration of each customer's most recent call and the number of calls within the historical statistical period based on the customer service call data of each customer in the target customer service call data;
[0014] The value group classification module is used to perform binary classification on the duration of the most recent call, the number of calls and the average value of the content complexity of each customer in the historical statistical period based on the RFM model according to a preset threshold, and obtain the value group classification results of all calling customers, so as to allocate caller seats to the customer calls according to the value group classification results when receiving customer calls.
[0015] In a third aspect, an embodiment of the present application provides an electronic device, including:
[0016] A processor, a memory, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any of the above-mentioned methods for allocating incoming call seats is implemented.
[0017] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium. When instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute any of the above-mentioned methods for allocating incoming call seats.
[0018] Compared with the prior art, the embodiments of the present application have the following advantages:
[0019] In an embodiment of the present application, target customer service call data of a specified business scenario is screened out from customer service call data within a historical statistical period from the current time. The pre-trained content complexity prediction model is called to process each customer service call data of each customer in the target customer service call data to obtain the content complexity of each customer's call each time. The content complexity is predicted by the content complexity prediction model based on the call business type and call emotion information of each customer at each call. According to the customer service call data of each customer in the target customer service call data, the duration of each customer's most recent call and the number of calls within the historical statistical period are determined. Based on the RFM model, the duration of the most recent call, the number of calls, and the average value of the content complexity of each customer within the historical statistical period are binary classified according to the preset threshold value to obtain the value group classification results of all calling customers, so that when receiving customer calls, caller seats are allocated to customer calls according to the value group classification results. The embodiment of the present application uses a content complexity model to fully explore the complexity of customer call content, and combines the RFM model to classify customer value groups, thereby achieving efficient and reasonable allocation of agent resources and ensuring that high-quality services can still be provided during peak hours.
[0020] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Figure 1 A flowchart of a method for allocating seats for incoming calls provided in an embodiment of the present application;
[0022] Figure 2 A schematic diagram of a seat resource allocation process provided in an embodiment of the present application;
[0023] Figure 3 A schematic diagram of a content complexity model training process provided in an embodiment of the present application;
[0024] Figure 4 A schematic diagram of the structure of a device for allocating seats for incoming calls provided in an embodiment of the present application;
[0025] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0026] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.
[0027] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms "a", "said" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.
[0028] Reference Figure 1 , shows a flowchart of the steps of a method for allocating seats for incoming calls provided by an embodiment of the present application, such as Figure 1 As shown, the method for allocating seats for incoming calls may include: step 101, step 102, step 103 and step 104.
[0029] Step 101: Filter out target customer service call data of a specified business scenario from customer service call data within a historical statistical period from the current time.
[0030] In this embodiment, the historical statistical period may be a period for counting customer service call data, which is one month or three months away from the current time.
[0031] When grouping incoming callers, customer service call data from a customer service platform (such as a bank platform, etc.) within a historical statistical period from the current time can be obtained.
[0032] After obtaining the customer service call data within the historical statistical period, you can filter out the target customer service call data for the specified business scenario from these customer service call data. For example, you can count all customer service calls on the bank platform in the past year, and then filter the call data to only retain call information involving key business scenarios such as consultation, complaints, and suggestions.
[0033] After the target customer service call data of the specified business scenario is screened out from the customer service call data within the historical statistical period from the current time, step 102 and step 103 are executed.
[0034] Step 102: Call a pre-trained content complexity prediction model to process each customer service call data of each customer in the target customer service call data to obtain the content complexity of each customer's call each time; the content complexity is predicted by the content complexity prediction model based on the call business type and call emotion information of each customer at each call.
[0035] The content complexity prediction model can be used to predict the complexity of customer call content. The higher the content complexity, the more complex the corresponding business is, and it needs to be handed over to experienced agents for processing.
[0036] The training process of the content complexity prediction model can be described in detail in combination with the following specific implementation methods.
[0037] In a specific implementation of the present application, before the above step 102, the following may also be included:
[0038] Step A1: segment the historical incoming call data according to the semantic completeness to obtain segmented incoming call data corresponding to the historical incoming call data.
[0039] In this embodiment, historical call data may be obtained when training the content complexity prediction model. In this example, the historical call data may be extracted from a database of a designated customer service platform or obtained from the Internet. This embodiment does not limit the method for obtaining the historical call data.
[0040] After obtaining the historical call data, the historical call data can be segmented according to semantic completeness to obtain segmented call data corresponding to the historical call data. Specifically, the customer call content can be segmented according to semantic completeness to ensure that each segment has independent business significance, and the segmented content can be scored by model respectively.
[0041] After the historical incoming call data is segmented according to the semantic completeness to obtain the segmented incoming call data corresponding to the historical incoming call data, step A2 is executed.
[0042] Step A2: performing identification processing on the segmented incoming call data to obtain segmented service types and segmented emotion information of the segmented incoming call data.
[0043] After the historical call data is segmented according to the semantic completeness to obtain the segmented call data corresponding to the historical call data, the segmented call data can be identified and processed to obtain the segmented business type and segmented emotional information of the segmented call data. Specifically, each segmented call data can be identified by type, such as consultation, complaint, suggestion, etc., so as to better understand the business type to which the call content belongs. Different business types have different levels of complexity, and extracting this feature can effectively identify the complexity of the content. At the same time, the emotional tendency in each segmented call data can be identified, such as positive, negative, neutral, etc. Different emotional tendencies can indirectly reflect the complexity of the segmented content, and extracting this feature can effectively identify the complexity of the content.
[0044] After the segmented incoming call data is identified and processed to obtain the segmented service types and segmented emotion information of the segmented incoming call data, step A3 is performed.
[0045] Step A3: Obtain Bigram features and TF-IDF features of the segmented incoming call data.
[0046] After the segmented incoming call data is identified and processed to obtain the segmented service type and segmented emotion information of the segmented incoming call data, the Bigram features and TF-IDF features of the segmented incoming call data can be obtained. The implementation process can be described in detail in combination with the following specific implementation methods.
[0047] In another specific implementation of the present application, the above sub-step A3 may include:
[0048] Sub-step B1: Obtaining bigrams of the segmented incoming call data.
[0049] In this embodiment, after obtaining the segmented incoming call data, a bigram of the segmented incoming call data can be obtained, where a bigram refers to two consecutive words in a text as a phrase. In a specific implementation, key information to be analyzed can be extracted from the segmented incoming call data, and the extracted information can be arranged in order, and two consecutive information items can be taken as a bigram.
[0050] After obtaining the bigram phrases of the segmented incoming call data, sub-step B2 is performed.
[0051] Sub-step B2: Encode the bigram to obtain vector encoding features of the bigram.
[0052] After obtaining the bigrams of the segmented incoming call data, the bigrams can be encoded to obtain vector encoding features of the bigrams. Specifically, Word2Vec can be used to perform unsupervised training on the bigrams to generate a dense vector, i.e., a vector encoding feature, for each bigram.
[0053] After encoding the bigram to obtain the vector encoding features of the bigram, sub-step B3 is executed.
[0054] Sub-step B3: performing a value averaging operation on the vector encoding features of all bigrams of the segmented incoming call data to obtain the Bigram features of the segmented incoming call data.
[0055] After encoding the bigrams to obtain the vector encoding features of the bigrams, the vector encoding features of all the bigrams in the segmented incoming call data can be averaged to obtain the Bigram features of the segmented incoming call data. Specifically, AVG-Pooling can be performed on the dense vectors of all the bigrams in each segmented content to extract the overall semantic features (i.e., Bigram features), which can capture more contextual information than a single word.
[0056] Sub-step B4: Calculate the word frequency of each of the bigrams in the segmented incoming call data, and the inverse document frequency of each of the bigrams in the word incoming calls corresponding to the segmented incoming call data.
[0057] After obtaining the bigrams of the segmented incoming call data, the word frequency of each bigram in the segmented incoming call data and the inverse document frequency of each bigram in the word incoming calls corresponding to the segmented incoming call data may also be calculated.
[0058] Sub-step B5: Determine the TF-IDF features of the segmented incoming call data according to the word frequency and the inverse document frequency.
[0059] Then, the TF-IDF features of the segmented incoming call data can be determined based on the word frequency and inverse document frequency. Specifically, for each bigram, its word frequency (TF) in the current segment and its inverse document frequency (IDF) in the entire incoming call are calculated, and then the TF-IDF value of the word in the current segment is calculated. Each segmented content is converted into a TF-IDF vector feature, where each dimension corresponds to the TF-IDF value of a bigram, and bigrams with small TF-IDF values are removed to reduce the problem of vector dimensions being too large to affect training.
[0060] After obtaining the Bigram features and TF-IDF features of the segmented incoming call data, step A4 is performed.
[0061] Step A4: constructing a model training sample based on the Bigram feature, the TF-IDF feature, the segmented business type and the segmented emotion information.
[0062] After obtaining the Bigram features and TF-IDF features of the segmented incoming call data, the model training samples can be constructed based on the Bigram features, TF-IDF features, segmented business types and segmented emotional information. Specifically, the segmented business types are encoded (such as one-hot encoding, etc.) to obtain segmented type features. And the segmented emotional information is encoded (such as one-hot encoding, etc.) to obtain segmented emotional features. Therefore, the model training samples can be constructed based on the Bigram features, TF-IDF features, segmented type features and segmented emotional features.
[0063] Step A5: Determine the content complexity label of the segmented incoming call data according to the label complexity, call duration and call rate of the segmented incoming call data.
[0064] After obtaining the segmented incoming call data, the content complexity label of the segmented incoming call data can be determined according to the label complexity, call duration and call rate of the segmented incoming call data. In this embodiment, the target variable (i.e., content complexity label) of the content complexity model can be defined by weighted fusion of indicators such as "artificial complexity label" (i.e., label complexity in this example), "call duration" and "call rate".
[0065] Among them, manual complexity marking: experienced agents can score the complexity of the segmented content, with the score range from 0 to 1, and the larger the value, the more complex the content.
[0066] Introducing call duration and call rate: In order to reduce the error of manual labeling, we combine the two objective indicators of call duration and call rate to indirectly characterize the complexity of the call content:
[0067] Call duration: Calculate the call duration of each segment. The longer the duration, the more complex the content. However, this may also be affected by the call speed. Therefore, duration is only a reference factor for complexity assessment.
[0068] Talk rate: At the same time, the average speaking rate of each talk segment is calculated to distinguish between scenes with slow speaking speed but simple content and scenes with fast speaking speed but complex content. It is recorded as rate = number of words in segment content / segment time.
[0069] Construction of target variables: The artificial complexity mark, call duration and call rate can be weighted and fused to form a content complexity label, which is used to train the content complexity prediction model.
[0070] In one implementation of the present application, the complexity weight of the pre-set marking complexity, the duration weight of the call duration, and the rate weight of the call rate can be obtained. Based on the complexity weight, duration weight, and rate weight, the marking complexity, call duration, and call rate are weighted and summed to obtain the content complexity label of the segmented incoming call data. The calculation formula can be as follows:
[0071] label=w1·arti+w2·time+w3·rate
[0072] In the above formula, label is the content complexity label, w1 is the complexity weight, w2 is the duration weight, w3 is the rate weight, arti is the marking complexity, time is the call duration, and rate is the call rate.
[0073] Step A6: The content complexity prediction model is obtained based on the model training samples and the content complexity labels.
[0074] After obtaining the model training samples and the content complexity labels, a content complexity prediction model can be obtained by training based on the model training samples and the content complexity labels. The model training process can be described in detail in conjunction with the following specific implementation methods.
[0075] In another specific implementation of the present application, the above step A6 may include:
[0076] Sub-step C1: inputting the model training sample and the content complexity label into the content complexity prediction model to be trained, so that the model training sample is processed by the content complexity prediction model to be trained.
[0077] In this embodiment, after obtaining the model training sample and the content complexity label, the model training sample and the content complexity label can be input into the content complexity prediction model to be trained. The model training sample is processed by the content complexity prediction model to be trained. The content complexity prediction model to be trained can be a model of a neural network, a decision tree, a random forest or other machine learning algorithms.
[0078] Sub-step C2: obtaining the predicted content complexity score corresponding to the segmented incoming call data output by the content complexity prediction model to be trained.
[0079] The content complexity prediction model to be trained can process each segmented incoming call data and output a predicted content complexity score, which represents the complexity of the segmented content as considered by the model.
[0080] Sub-step C3: Based on the content complexity label and the predicted content complexity score with the maximum score value, calculate the loss value of the content complexity prediction model to be trained.
[0081] The loss value is calculated using the content complexity label (true value) and the predicted content complexity score of each segmented call data. The loss value is a metric that measures the accuracy of the model's predictions, which is usually calculated by comparing the difference between the predicted value and the true value.
[0082] Sub-step C4: when the loss value is within a preset range, obtaining the final content complexity prediction model.
[0083] Check whether the calculated loss value is within the preset range. This range is set according to the requirements and represents the acceptable size of the model prediction error.
[0084] If the loss value is within the preset range, the model can be considered to have achieved sufficient accuracy and used as the final content complexity prediction model. If the loss value is not within the preset range, the model needs to be further adjusted and optimized, and then the above steps are repeated until the requirements are met.
[0085] The training process of the content complexity prediction model can be as follows: Figure 3 As shown, the training process of the content complexity prediction model can include four modules: 1. call content preprocessing, 2. feature processing, 3. target definition, and 4. model training.
[0086] 1. Call content preprocessing: Use the pre-trained large language model to preprocess the call content, including the following three aspects:
[0087] Call content segmentation: Segment customer call content according to semantic integrity to ensure that each segment has independent business significance, and perform model scoring on the segmented content separately.
[0088] Segment type identification: Identify the type of each segment, such as consultation, complaint, suggestion, etc., to better understand the business type to which the call content belongs. Different business types have different levels of complexity, and extracting this feature can effectively identify the complexity of the content.
[0089] Incoming call emotion recognition: Identify the emotional tendency in each segment, such as positive, negative, neutral, etc. Different emotional tendencies can indirectly reflect the complexity of the segmented content. Extracting this feature can effectively identify the complexity of the content.
[0090] 2. Feature processing: For the pre-processed segmented content, feature processing is completed through TF-IDF, Bigram and other technologies:
[0091] Bigram features: Generate Bigrams (bigrams) for all incoming call content, use Word2Vec to perform unsupervised training on the bigrams, generate dense vectors for each bigram, perform AVG-Pooling on the dense vectors of all bigrams in each segmented content, and extract overall semantic features, which can capture more contextual information than a single word.
[0092] TF-IDF vector feature: For each bigram, calculate its term frequency (TF) in the current segment and its inverse document frequency (IDF) in the entire call, and then calculate the TF-IDF value of the term in the current segment. Convert each segmented content into a TF-IDF vector feature, where each dimension corresponds to the TF-IDF value of a bigram. Remove bigrams with small TF-IDF values to reduce the problem of vector dimensions being too large to affect training.
[0093] Type feature: Based on the pre-trained large prediction model, the business type of the segmented content is identified, one-hot encoded, and a sparse vector is formed as the input variable.
[0094] Emotional features: Based on the pre-trained large prediction model, the emotional tendency of the segmented content is identified, one-hot encoded, and a sparse vector is formed as the input variable.
[0095] 3. Target definition: By weighted fusion of indicators such as "artificial complexity labeling", "call duration" and "call rate", the target variable of the content complexity model is defined.
[0096] Manual complexity marking: Experienced agents are invited to rate the complexity of the segmented content, with a rating range of 0-1. The larger the value, the more complex the content. This is denoted as arti.
[0097] Introducing call duration and call rate: In order to reduce the error of manual labeling, the two objective indicators of call duration and call rate are combined to indirectly characterize the complexity of the call content.
[0098] Call duration: Calculate the call duration of each segment. The longer the duration, the more complex the content may be; however, this may also be affected by the call speed. Therefore, the duration is only a reference factor for complexity assessment. Recorded as time.
[0099] Talk rate: At the same time, the average speaking rate of each talk segment is calculated to distinguish between scenes with slow speaking speed but simple content and scenes with fast speaking speed but complex content. It is recorded as rate = number of words in segment content / segment time.
[0100] Constructing target variables: The artificial complexity label, call duration, and call rate can be weighted and fused to form a content complexity label, which is used to train the content complexity prediction model.
[0101] 4. Model training:
[0102] Model input: The input of the content complexity model includes Bigram features, TF-IDF vector features, type features, and sentiment features.
[0103] Model structure: The LightGBM algorithm is used to train a content complexity model based on sample data to predict the complexity of incoming call content. LightGBM is a gradient boosting framework based on a decision tree algorithm, suitable for processing large-scale data sets; the algorithm reduces the amount of computation and memory usage while increasing the training speed by using a histogram-based algorithm and leaf node mutual exclusion constraints.
[0104] Model output: Output the content complexity score for each segmented call content; since a call is divided into multiple segments, the content complexity scores of multiple segments are compared, and the highest score is selected as the content complexity score of the call.
[0105] After obtaining the content complexity prediction model, the pre-trained content complexity prediction model can be called to process each customer service call data of each customer in the target customer service call data to obtain the content complexity of each customer's call. The content complexity can be predicted by the content complexity prediction model based on the call service type and call emotion information of each customer at each call.
[0106] The prediction process of the content complexity of each incoming call can be:
[0107] Each customer service call data is segmented according to semantic completeness to obtain multiple segmented call data of each customer service call data, and the call business type and call emotion information of the multiple segmented call data are obtained. The call business type and call emotion information of the multiple segmented call data are encoded to obtain the encoding features as the input of the content complexity prediction model.
[0108] The content complexity prediction model can process the encoding features to output predicted content complexity scores for multiple segmented call data, and then the predicted content complexity score with the largest score value can be screened out from the multiple predicted content complexity scores as the final content complexity score, that is, the predicted content complexity score with the largest score value will be used as the content complexity of each customer service call data.
[0109] Step 103: Determine the duration of each customer's most recent call and the number of calls within the historical statistical period based on the customer service call data of each customer in the target customer service call data.
[0110] After obtaining the target customer service call data, the customer service call data of each customer in the target customer service call data can be obtained, and the customer service call data of each customer in the target customer service call data can be used to determine the duration of each customer's most recent call and the number of calls in the historical statistical period. Specifically, the call data of each customer can be summarized to calculate the number of days (Recency) from the last call to the statistical time point, the number of calls in the statistical period (Frequency), and the average value (Monetary) of the call content complexity score in the statistical period, which can be used as the input variables of the RFM model.
[0111] Step 104: Based on the RFM model, the duration of the most recent call, the number of calls and the average value of the content complexity of each customer in the historical statistical period are binary classified according to a preset threshold value to obtain the value group classification results of all calling customers, so that when a customer call is received, the customer call is assigned a caller seat according to the value group classification results.
[0112] Then, based on the RFM model, the duration of the most recent call, the number of calls, and the average value of the content complexity of each customer in the historical statistical period can be binary classified according to the preset threshold, and the value group classification results of all calling customers can be obtained, so that when receiving customer calls, call agents can be assigned to customer calls according to the value group classification results. Caller agents, also known as operators or agent representatives, are the basic units in call centers that handle fax, call, email and other services.
[0113] In the specific implementation, the thresholds of the three dimensions of Recency, Frequency, and Monetary can be set for binary classification (high, low) through training, so as to divide the historical call customers into eight different value groups. The thresholds of the three dimensions correspond to the thresholds corresponding to the average value of the duration of the most recent call, the number of calls, and the content complexity of each customer in the historical statistical period. The eight different value groups can be shown in the following Table 1:
[0114] Table 1: Classification of value groups
[0115]
[0116] In Table 1 above, a high Recency value indicates that the most recent call was close to the statistical time point; a high Frequency value indicates that there are many calls within the statistical period; and a high Monetary value indicates that the call content is highly complex.
[0117] The RFM model divides incoming calls into eight different value groups. The higher up the table goes, the higher the value of the response group. The so-called high value refers to customers whose historical calls are complex and who are in urgent need of solving problems. These customer calls require more professional agents (high-performing agents, team leaders) to solve customer problems as soon as possible and improve customer experience. Ordinary agents cannot efficiently solve these high-value customer calls in a short time, but ordinary agents can effectively solve low-value customer calls. Through RFM grouping, the customer service team can optimize resource allocation and improve overall service quality and efficiency.
[0118] In this embodiment, the thresholds of the three dimensions can also be continuously updated over time. Specifically, in order to maximize the distinction between the eight value groups, the variance of the probability of customers calling in the next 30 days is used as the objective function, and the Bayesian optimization algorithm is used to find the optimal parameter settings, or parameters are specified according to business needs.
[0119] In this embodiment, in order to overcome the limitation of manually arranging seats and the inability to achieve global optimization, the operations research ILP (Integer Linear Programming) model is introduced to achieve an overall reduction in the proportion of secondary calls (second calls). Under the constraints of the number of online seats, the allocation of non-negative numbers, and the integer type of the allocated seats, the overall optimization is achieved by minimizing the objective function. The implementation process can be described in detail in conjunction with the following specific implementation methods.
[0120] In a specific implementation of the present application, after the above step 104, the following may also be included:
[0121] Step D1: upon receiving a call from a target customer, determining the target value group category to which the target customer belongs according to the value group classification result.
[0122] In this embodiment, when a call is received from a target customer, the target value group category to which the target customer belongs can be determined based on the value group classification result. For example, the value group category can be queried from Table 1 above based on the customer identification of the target customer to obtain the target value group category to which the target customer belongs.
[0123] After determining the target value group category to which the target customer belongs based on the value group classification result, step D2 is executed.
[0124] Step D2: allocating caller seats to the target customers based on the ILP model of operations research and the quantity restriction conditions corresponding to the target value group category and the caller seat type.
[0125] After determining the target value group category to which the target customer belongs according to the value group classification result, caller seats can be allocated to the target customer based on the operations research ILP model according to the target value group category and the quantity restriction conditions corresponding to the caller seat type.
[0126] In a specific implementation, the ILP model may include:
[0127] 1. Define variables: set x ij is a decision variable, indicating the number of customers of type i assigned to type j agents, where i represents the RFM group and j represents the agent type (ordinary agent, high-performance agent, team leader).
[0128] 2. Objective function: The goal is to minimize the overall two-electric ratio, which can be expressed as:
[0129]
[0130] In the above formula, c ij It is the second-order call ratio associated with the i-th customer being served by the j-th agent.
[0131] 3. Constraints:
[0132] (1) Seat quantity limit: Ensure that the number of customers assigned to each type of seat does not exceed the maximum number of available online seats S j :
[0133]
[0134] (2) Non-negativity constraint: The number of seats allocated must be non-negative:
[0135] x ij ≥0for all i,j
[0136] (3) Integer constraint: The number of seats allocated must be an integer:
[0137] x ij ∈Z for all i,j
[0138] Z represents the set of integers.
[0139] Through the objective function and constraints set above, historical call data c is counted ij , the planned online seat capacity for the next day, using the solver to calculate x for the next day ij , can effectively assign customer groups to different types of seats. This approach not only helps to reduce the overall second-tier ratio, but also improves customer satisfaction and service efficiency.
[0140] When a customer calls, the agent system will send the customer information to the online agent configuration strategy module. The module will query the RFM customer group to which the historical stock customers belong, determine that the current customer belongs to customer group i, and calculate the weight of customer group i in the jth agent to which it belongs:
[0141]
[0142] The assigned agent type is determined based on weighted random sampling:
[0143] Choose k if:
[0144] Among them, r is a random number uniformly distributed in the interval [0,1]. When r falls within the range of the above formula, k types of seats are selected and returned to the seat system.
[0145] The method for allocating call seats provided in the embodiment of the present application is to filter out target customer service call data of a specified business scenario from customer service call data within a historical statistical period from the current time. The pre-trained content complexity prediction model is called to process each customer service call data of each customer in the target customer service call data to obtain the content complexity of each customer's call each time. The content complexity is predicted by the content complexity prediction model based on the call business type and call emotion information of each customer at each call. According to the customer service call data of each customer in the target customer service call data, the duration of each customer's most recent call and the number of calls in the historical statistical period are determined. Based on the RFM model, the duration of the most recent call, the number of calls, and the average value of the content complexity of each customer in the historical statistical period are binary classified according to a preset threshold value to obtain the value group classification results of all calling customers, so that when receiving a customer call, the call seat is allocated to the customer call according to the value group classification result. The embodiment of the present application uses a content complexity model to fully explore the complexity of customer call content, and combines the RFM model to classify customer value groups, thereby achieving efficient and reasonable allocation of agent resources and ensuring that high-quality services can still be provided during peak hours.
[0146] Reference Figure 4 , which shows a schematic diagram of the structure of a device for allocating seats for incoming calls provided in an embodiment of the present application, such as Figure 4 As shown, the incoming call agent allocation device 400 may include the following modules:
[0147] The target call data screening module 410 is used to screen the target customer service call data of the specified business scenario from the customer service call data within the historical statistical period from the current time;
[0148] The content complexity acquisition module 420 is used to call the pre-trained content complexity prediction model to process each customer service call data of each customer in the target customer service call data to obtain the content complexity of each customer's call each time; the content complexity is predicted by the content complexity prediction model based on the call service type and call emotion information of each customer at each call;
[0149] The call count determination module 430 is used to determine the duration of the most recent call of each customer and the number of calls within the historical statistical period according to the customer service call data of each customer in the target customer service call data;
[0150] The value group classification module 440 is used to perform binary classification on the duration of the most recent call, the number of calls and the average value of the content complexity of each customer in the historical statistical period based on the RFM model according to a preset threshold, and obtain the value group classification results of all calling customers, so as to allocate caller seats to the customer calls according to the value group classification results when receiving customer calls.
[0151] Optionally, the device further comprises:
[0152] A segmented incoming call data acquisition module is used to segment the historical incoming call data according to the semantic completeness to obtain the segmented incoming call data corresponding to the historical incoming call data;
[0153] A segment information acquisition module, used to identify and process the segmented incoming call data to obtain segmented business types and segmented emotion information of the segmented incoming call data;
[0154] A segmented incoming call feature acquisition module, used to acquire Bigram features and TF-IDF features of the segmented incoming call data;
[0155] A model sample construction module, used to construct a model training sample based on the Bigram feature, the TF-IDF feature, the segmented business type and the segmented emotion information;
[0156] A complexity label determination module, used to determine the content complexity label of the segmented incoming call data according to the marking complexity, call duration and call rate of the segmented incoming call data;
[0157] The complexity prediction model training module is used to obtain the content complexity prediction model based on the model training samples and the content complexity labels.
[0158] Optionally, the segmented incoming call feature acquisition module includes:
[0159] A bigram acquisition unit, used for acquiring bigrams of the segmented incoming call data;
[0160] A coding feature acquisition unit, used to perform coding processing on the bigram to obtain a vector coding feature of the bigram;
[0161] A Bigram feature acquisition unit, configured to perform a value averaging operation on the vector encoding features of all the bigrams of the segmented incoming call data to obtain the Bigram features of the segmented incoming call data;
[0162] An inverse document frequency acquisition unit, used for calculating the word frequency of each of the bigrams in the segmented incoming call data, and the inverse document frequency of each of the bigrams in the word calls corresponding to the segmented incoming call data;
[0163] The TF-IDF feature determination unit is used to determine the TF-IDF feature of the segmented incoming call data according to the word frequency and the inverse document frequency.
[0164] Optionally, the model sample construction module includes:
[0165] A type feature acquisition unit, used to encode the segment service type to obtain a segment type feature;
[0166] An emotion feature acquisition unit, used for encoding the segmented emotion information to obtain segmented emotion features;
[0167] The model sample construction unit is used to construct the model training sample based on the Bigram feature, the TF-IDF feature, the segment type feature and the segment emotion feature.
[0168] Optionally, the complexity label determination module includes:
[0169] A weight acquisition unit, used to acquire a preset complexity weight of the marking complexity, a duration weight of the call duration, and a rate weight of the call rate;
[0170] The complexity label acquisition unit is used to perform weighted summation of the marking complexity, the call duration and the call rate based on the complexity weight, the duration weight and the rate weight to obtain the content complexity label of the segmented incoming call data.
[0171] Optionally, the complexity prediction model training module includes:
[0172] A model sample input unit, used for inputting the model training sample and the content complexity label into the content complexity prediction model to be trained, so that the model training sample is processed by the content complexity prediction model to be trained;
[0173] A prediction complexity score acquisition unit, used to acquire the prediction content complexity score corresponding to the segmented incoming call data output by the content complexity prediction model to be trained;
[0174] A loss value calculation unit, configured to calculate a loss value of the content complexity prediction model to be trained based on the content complexity label and the predicted content complexity score;
[0175] The complexity prediction model acquisition unit is used to obtain the final content complexity prediction model when the loss value is within a preset range.
[0176] Optionally, the device further comprises:
[0177] A target group category determination module is used to determine the target value group category to which the target customer belongs based on the value group classification result when receiving a call from the target customer;
[0178] The caller agent allocation module is used to allocate caller agents to the target customers based on the operations research ILP model according to the target value group category and the quantity restriction conditions corresponding to the caller agent type.
[0179] The call seat allocation device provided in the embodiment of the present application selects the target customer service call data of the specified business scenario from the customer service call data within the historical statistical period from the current time. The pre-trained content complexity prediction model is called to process each customer service call data of each customer in the target customer service call data to obtain the content complexity of each customer's call each time. The content complexity is predicted by the content complexity prediction model based on the call business type and call emotion information of each customer at each call. According to the customer service call data of each customer in the target customer service call data, the duration of each customer's most recent call and the number of calls in the historical statistical period are determined. Based on the RFM model, the duration of the most recent call, the number of calls, and the average value of the content complexity of each customer in the historical statistical period are binary classified according to the preset threshold value to obtain the value group classification results of all calling customers, so that when receiving customer calls, call seats are allocated to customer calls according to the value group classification results. The embodiment of the present application uses a content complexity model to fully explore the complexity of customer call content, and combines the RFM model to classify customer value groups, thereby achieving efficient and reasonable allocation of agent resources and ensuring that high-quality services can still be provided during peak hours.
[0180] An embodiment of the present application further provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program implements the above-mentioned caller agent allocation method when executed by the processor.
[0181] Figure 5 FIG. 5 is a schematic diagram showing the structure of an electronic device 500 according to an embodiment of the present invention. Figure 5As shown, the electronic device 500 includes a central processing unit (CPU) 501, which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) 502 or computer program instructions loaded from a storage unit 508 into a random access memory (RAM) 503. In RAM 503, various programs and data required for the operation of the electronic device 500 can also be stored. CPU 501, ROM 502 and RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.
[0182] Multiple components in the electronic device 500 are connected to the I / O interface 505, including: an input unit 506, such as a keyboard, a mouse, a microphone, etc.; an output unit 507, such as various types of displays, speakers, etc.; a storage unit 508, such as a disk, an optical disk, etc.; and a communication unit 509, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 509 allows the electronic device 500 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.
[0183] The various processes and processing described above may be performed by the processing unit 501. For example, the method of any of the above embodiments may be implemented as a computer software program, which is tangibly contained in a computer-readable medium, such as the storage unit 508. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 500 via the ROM 502 and / or the communication unit 509. When the computer program is loaded into the RAM 503 and executed by the CPU 501, one or more actions in the method described above may be performed.
[0184] In addition, an embodiment of the present application further provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the above-mentioned caller agent allocation method is implemented.
[0185] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.
[0186] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, devices, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0187] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, terminals (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0188] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0189] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal so that a series of operating steps are executed on the computer or other programmable terminal to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable terminal. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0190] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present application.
[0191] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or terminal including the elements.
[0192] The above is a detailed introduction to a method for allocating incoming call seats, a device for allocating incoming call seats, an electronic device and a computer-readable storage medium provided by the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for a person skilled in the art, according to the idea of the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.
Claims
1. A method for allocating seats for incoming calls, characterized in that: The method comprises: Filter the target customer service call data of the specified business scenario from the customer service call data within the historical statistical period from the current time; Calling a pre-trained content complexity prediction model to process each customer service call data of each customer in the target customer service call data to obtain the content complexity of each customer's call each time; the content complexity is predicted by the content complexity prediction model based on the call service type and call emotion information of each customer at each call; Determine the duration of each customer's most recent call and the number of calls within the historical statistical period based on the customer service call data of each customer in the target customer service call data; Based on the RFM model, the duration of the most recent call, the number of calls and the average value of the content complexity of each customer in the historical statistical period are binary classified according to a preset threshold, and the value group classification results of all calling customers are obtained, so that when a customer call is received, the customer call is assigned a caller seat according to the value group classification results.
2. The method according to claim 1, characterized in that: Before calling the pre-trained content complexity prediction model to process each customer service call data of each customer in the target customer service call data to obtain the content complexity of each call of each customer, the method further includes: Segmenting the historical incoming call data according to semantic completeness to obtain segmented incoming call data corresponding to the historical incoming call data; Identify and process the segmented incoming call data to obtain segmented service types and segmented emotion information of the segmented incoming call data; Obtaining Bigram features and TF-IDF features of the segmented incoming call data; Constructing a model training sample based on the Bigram feature, the TF-IDF feature, the segmented business type and the segmented emotion information; Determining a content complexity label of the segmented incoming call data according to the label complexity, call duration, and call rate of the segmented incoming call data; The content complexity prediction model is obtained by training based on the model training samples and the content complexity labels.
3. The method according to claim 2, characterized in that The obtaining of Bigram features and TF-IDF features of the segmented incoming call data includes: Obtaining a bigram phrase of the segmented incoming call data; Performing encoding processing on the bigram to obtain vector encoding features of the bigram; Performing a value averaging operation on the vector encoding features of all bigrams of the segmented incoming call data to obtain the Bigram features of the segmented incoming call data; Calculate the word frequency of each bigram in the segmented incoming call data, and the inverse document frequency of each bigram in the word incoming calls corresponding to the segmented incoming call data; The TF-IDF features of the segmented incoming call data are determined according to the word frequency and the inverse document frequency.
4. The method according to claim 2, characterized in that: The constructing of model training samples based on the Bigram feature, the TF-IDF feature, the segmented business type and the segmented emotion information includes: Encoding the segmented service type to obtain segmented type characteristics; Encoding the segmented emotion information to obtain segmented emotion features; The model training sample is constructed based on the Bigram feature, the TF-IDF feature, the segment type feature and the segment emotion feature.
5. The method according to claim 2, characterized in that: The determining of the content complexity label of the segmented incoming call data according to the label complexity, call duration and call rate of the segmented incoming call data includes: Obtaining a preset complexity weight of the marking complexity, a duration weight of the call duration, and a rate weight of the call rate; The tag complexity, the call duration and the call rate are weighted and summed based on the complexity weight, the duration weight and the rate weight to obtain a content complexity label of the segmented incoming call data.
6. The method according to claim 2, characterized in that The step of obtaining the content complexity prediction model based on the model training sample and the content complexity label training includes: Inputting the model training sample and the content complexity label into a content complexity prediction model to be trained, so that the model training sample is processed by the content complexity prediction model to be trained; Obtaining a predicted content complexity score corresponding to the segmented incoming call data output by the content complexity prediction model to be trained; Based on the content complexity label and the predicted content complexity score, a loss value of the content complexity prediction model to be trained is calculated; When the loss value is within a preset range, the final content complexity prediction model is obtained.
7. The method according to claim 1, characterized in that After the RFM model is used to perform binary classification on the duration of the most recent call, the number of calls, and the average value of the content complexity of each customer in the historical statistical period according to a preset threshold value to obtain the value group classification results of all calling customers, the method further includes: When receiving a call from a target customer, determining the target value group category to which the target customer belongs according to the value group classification result; Based on the operations research ILP model, caller seats are allocated to the target customers according to the target value group category and the quantity restriction conditions corresponding to the caller seat type.
8. A device for allocating seats for incoming calls, characterized in that: The device comprises: A target call data screening module is used to screen target customer service call data for a specified business scenario from customer service call data within a historical statistical period from the current time; A content complexity acquisition module is used to call a pre-trained content complexity prediction model to process each customer service call data of each customer in the target customer service call data to obtain the content complexity of each customer's call each time; the content complexity is predicted by the content complexity prediction model based on the call service type and call emotion information of each customer at each call; A call count determination module, used to determine the duration of each customer's most recent call and the number of calls within the historical statistical period based on the customer service call data of each customer in the target customer service call data; The value group classification module is used to perform binary classification on the duration of the most recent call, the number of calls and the average value of the content complexity of each customer in the historical statistical period based on the RFM model according to a preset threshold, and obtain the value group classification results of all calling customers, so as to allocate caller seats to the customer calls according to the value group classification results when receiving customer calls.
9. An electronic device, characterized in that: include: A processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method for assigning incoming call seats according to any one of claims 1 to 7 when executing the program.
10. A computer-readable storage medium, characterized in that: When the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the method for assigning call seats according to any one of claims 1 to 7.