Customer service allocation method, customer service index model training method, device and equipment

Abstract

The invention discloses a customer service allocation method, a customer service index model training method, a customer service index model training device and customer service index model training equipment. The method comprises the following steps: in response to a detected customer service allocation request of a to-be-detected user, obtaining at least one to-be-detected matching data combination; wherein the to-be-tested matching data combination comprises at least two kinds of to-be-tested feature data, and each piece of to-be-tested feature data comprises to-be-tested user data corresponding to a to-be-tested user and to-be-tested customer service data corresponding to a preset customer service; inputting each to-be-tested feature data in each to-be-tested matching data combination into a pre-trained target customer service index model to obtain output target customer service index data corresponding to preset customer service in each to-be-tested matching data combination; and obtaining a target customer service matched with the to-be-tested user from each preset customer service based on each target customer service index data. According to the embodiment of the invention, the problem that the user consultation demand is not matched with the manual customer service level is solved, and the customer service distribution accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of Internet technology, and in particular to a customer service allocation method, a customer service indicator model training method, a device and equipment. Background Art

[0002] With the development of Internet technology, new technologies such as big data, artificial intelligence, and 5G networks continue to emerge, and people's consumption patterns have also shifted from offline consumption to online shopping. In online shopping scenarios, customer service centers play an increasingly important role and their value positioning continues to evolve. It has become an important direction for the digital transformation of organizations such as governments and enterprises.

[0003] The traditional manual customer service center has a complex structure and relies solely on manual experience, which makes the contradiction between the growing online user consultation needs and the level of manual customer service increasingly prominent. Therefore, more and more online platforms have begun to use intelligent customer service robots to help manual customer service better serve customers.

[0004] In the process of implementing the present invention, it is found that there are at least the following technical problems in the prior art:

[0005] Limited by the capabilities of existing natural language understanding technology, intelligent customer service robots are far from being able to catch up with the service level of manual customer service and can only handle relatively simple questions. In addition, some users cannot accept the services of existing intelligent customer service robots. Therefore, how to achieve efficient and accurate matching of user consultation needs with the level of manual customer service is still a technical problem that needs to be solved urgently. Summary of the invention

[0006] The embodiments of the present invention provide a customer service allocation method, a training method, an apparatus and equipment for a customer service indicator model to solve the problem of mismatch between user consultation needs and manual customer service levels, improve the accuracy of customer service allocation, and thereby improve the service quality and order facilitation rate of customer service.

[0007] According to an embodiment of the present invention, a customer service allocation method is provided, the method comprising:

[0008] In response to detecting a customer service allocation request of a user to be tested, obtaining at least one set of matching data combinations to be tested; wherein the matching data combinations to be tested include at least two feature data to be tested, each of which includes user data to be tested corresponding to the user to be tested and customer service data to be tested corresponding to a preset customer service;

[0009] Input each feature data to be tested in each of the matching data combinations to be tested into a pre-trained target customer service indicator model, and obtain output target customer service indicator data corresponding to the preset customer services in each of the matching data combinations to be tested;

[0010] Based on the target customer service indicator data, a target customer service matching the user to be tested is obtained from the preset customer services.

[0011] According to another embodiment of the present invention, a method for training a customer service indicator model is provided, the method comprising:

[0012] Acquire at least one set of training sample data combination; wherein the training sample data combination includes at least two types of training feature data, each of the training feature data includes training user data corresponding to the training user and training customer service data corresponding to the training customer service;

[0013] Input each training feature data in each of the training sample data combinations into an untrained initial customer service indicator model to obtain output predicted customer service indicator data corresponding to the training customer service in each of the training sample data combinations;

[0014] Determine a training loss function based on each of the predicted customer service indicator data and the standard customer service indicator data corresponding to each of the training sample data combinations;

[0015] Based on the training loss function, the model parameters of the initial customer service indicator model are adjusted until the training loss function converges, and the initial customer service indicator model is used as the target customer service indicator model after training.

[0016] According to another embodiment of the present invention, a customer service allocation device is provided, the device comprising:

[0017] The module for acquiring the combination of matching data to be tested is used to acquire at least one group of combination of matching data to be tested in response to detecting a customer service allocation request of the user to be tested; wherein the combination of matching data to be tested includes at least two types of feature data to be tested, each of which includes user data to be tested corresponding to the user to be tested and customer service data to be tested corresponding to a preset customer service;

[0018] A target customer service indicator data output module is used to input each feature data to be tested in each matching data combination to a pre-trained target customer service indicator model, and obtain output target customer service indicator data corresponding to the preset customer service in each matching data combination to be tested;

[0019] The target customer service matching module is used to obtain the target customer service that matches the user to be tested from each of the preset customer services based on the target customer service indicator data.

[0020] According to another embodiment of the present invention, a training device for a customer service indicator model is provided, the device comprising:

[0021] A training sample data combination acquisition module is used to acquire at least one set of training sample data combinations; wherein the training sample data combination includes at least two types of training feature data, each of which includes training user data corresponding to a training user and training customer service data corresponding to a training customer service;

[0022] A predicted customer service indicator data output module, used to input each training feature data in each of the training sample data combinations into an untrained initial customer service indicator model, and obtain output predicted customer service indicator data corresponding to the training customer service in each of the training sample data combinations; a training loss function determination module, used to determine a training loss function based on each of the predicted customer service indicator data and the standard customer service indicator data corresponding to each of the training sample data combinations;

[0023] The target customer service indicator model determination module is used to adjust the model parameters of the initial customer service indicator model based on the training loss function until the training loss function converges, and then use the initial customer service indicator model as the trained target customer service indicator model.

[0024] According to another embodiment of the present invention, an electronic device is provided, the electronic device comprising:

[0025] at least one processor; and

[0026] a memory communicatively connected to the at least one processor; wherein,

[0027] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the customer service allocation method described in any embodiment of the present invention, and / or can execute the training method of the customer service indicator model described in any embodiment of the present invention.

[0028] According to another embodiment of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the customer service allocation method described in any embodiment of the present invention when executed, and / or to be able to execute the training method of the customer service indicator model described in any embodiment of the present invention.

[0029] The technical solution of the embodiment of the present invention, in response to detecting a customer service allocation request of a tested user, inputs at least two types of feature data to be tested in at least one group of matched data combinations to be tested into a pre-trained target customer service indicator model, respectively, to obtain output target customer service indicator data corresponding to the preset customer services in each matched data combination to be tested, and based on each target customer service indicator data, obtains the target customer service matching the tested user from each preset customer service, wherein each feature data to be tested includes the tested user data corresponding to the tested user and the tested customer service data corresponding to the preset customer service. The embodiment of the present invention constructs the correlation between the target customer service indicator data and the tested user data, solves the problem of mismatch between user consultation needs and manual customer service levels, improves the accuracy of subsequent customer service allocation, and thereby improves the service quality and order facilitation rate of customer service.

[0030] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0032] Figure 1 A flowchart of a customer service allocation method provided by an embodiment of the present invention;

[0033] Figure 2 A schematic diagram of a method for inputting a matching data combination to be tested provided by an embodiment of the present invention;

[0034] Figure 3 A flowchart of another customer service allocation method provided by an embodiment of the present invention;

[0035] Figure 4 A network architecture diagram of a multi-task weight unit provided by one embodiment of the present invention;

[0036] Figure 5 A network architecture diagram of a specific example of a target customer service indicator model provided by an embodiment of the present invention;

[0037] Figure 6 A flowchart of a method for training a customer service indicator model provided by one embodiment of the present invention;

[0038] Figure 7An architectural diagram of a specific example of a training sample data combination and standard customer service indicator data provided by an embodiment of the present invention;

[0039] Figure 8 A flowchart of another customer service indicator model training method provided by one embodiment of the present invention;

[0040] Fig. 9 A schematic diagram of a method for training a target text encoding module provided by an embodiment of the present invention;

[0041] Fig.10 A schematic diagram of the structure of a customer service allocation device provided by an embodiment of the present invention;

[0042] Fig.11 A schematic diagram of the structure of a training device for a customer service indicator model provided by one embodiment of the present invention;

[0043] Fig.12 The present invention is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0044] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0045] It should be noted that the terms "first", "second", "initial", "target", "reference", "training", "to be tested", "pre-training", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0046] Figure 1This is a flow chart of a customer service allocation method provided by an embodiment of the present invention. This embodiment can be applied to the situation of allocating customer service to reception users in a customer service reception scenario. The method can be executed by a customer service allocation device, which can be implemented in the form of hardware and / or software, and the customer service allocation device can be configured in a terminal device. Figure 1 As shown, the method includes:

[0047] S110 . In response to detecting a customer service assignment request of a user to be tested, obtaining at least one set of matching data combinations to be tested.

[0048] In this embodiment, the combination of matching data to be tested includes at least two types of feature data to be tested, and each feature data to be tested includes user data to be tested corresponding to the user to be tested and customer service data to be tested corresponding to the preset customer service. Specifically, the preset customer services corresponding to the customer service data to be tested in each combination of matching data to be tested are different. Exemplarily, the combination of matching data to be tested A includes the user data to be tested and the customer service data to be tested corresponding to the preset customer service A, and the combination of matching data to be tested B includes the user data to be tested and the customer service data to be tested corresponding to the preset customer service B.

[0049] Specifically, the user data to be tested can be used to characterize feature data related to the attributes of the user to be tested. Exemplarily, the user data to be tested includes but is not limited to user basic attributes and / or user behavior attributes. For example, user basic attributes may be gender, registration time, credit, etc., and user behavior attributes may be purchase volume, return rate, activity, etc. The user data to be tested is not limited here, and can be customized according to actual needs.

[0050] Specifically, the customer service data to be tested can be used to characterize the occupation-related characteristic data of the preset customer service. By way of example, the customer service data to be tested includes but is not limited to the department to which they belong, the number of days in the job, the historical reception volume and the average satisfaction, etc. The customer service data to be tested is not limited here, and can be customized according to actual needs.

[0051] In a specific embodiment, the combination of matching data to be tested also includes scene data to be tested and / or text data to be tested. Specifically, the scene data to be tested can be used to characterize feature data related to the conversation scene, and the text data to be tested can be used to characterize the conversation text between the user to be tested and the intelligent customer service robot or the previous manual customer service before the manual customer service is first assigned to the user to be tested or before the manual customer service is assigned again. Exemplarily, the scene data to be tested includes but is not limited to consultation item information, consultation intention, consultation order information, purchase time, return time, etc. The scene data to be tested is not limited here, and can be customized according to actual needs.

[0052] S120, inputting each feature data to be tested in each matching data combination to a pre-trained target customer service indicator model, and obtaining output target customer service indicator data corresponding to the preset customer service in each matching data combination to be tested.

[0053] In a specific embodiment, a serial processing strategy is adopted to input each feature data to be tested in each matching data combination to be tested into a pre-trained target customer service indicator model.

[0054] In another specific embodiment, the number of target customer service indicator models is multiple, and accordingly, a parallel processing strategy is adopted to input each feature data to be tested in each matching data combination to be tested into the corresponding target customer service indicator model. The number of target customer service indicator models is greater than or equal to the number of matching data combinations to be tested.

[0055] The advantage of this setting is that it improves the efficiency of obtaining target customer service indicator data, thereby improving the allocation efficiency of customer service allocation to meet the requirements for allocation efficiency in actual customer service allocation scenarios.

[0056] Figure 2 A schematic diagram of a method for inputting a matching data combination to be tested provided by an embodiment of the present invention. Specifically, Figure 2 The number of preset customer services shown is 4, namely customer service A, customer service B, customer service C and customer service D. Accordingly, based on the user data to be tested of the user to be tested and the customer service data to be tested corresponding to the four preset customer services, four matching data combinations to be tested are formed, and the four matching data combinations to be tested are respectively input into the target customer service indicator model A, the target customer service indicator model B, the target customer service indicator model C and the target customer service indicator model D. Specifically, the target customer service indicator model A outputs the target customer service indicator data corresponding to customer service A, the target customer service indicator model B outputs the target customer service indicator data corresponding to customer service B, the target customer service indicator model C outputs the target customer service indicator data corresponding to customer service C, and the target customer service indicator model D outputs the target customer service indicator data corresponding to customer service D.

[0057] In a specific embodiment, the target customer service indicator data includes at least one of target satisfaction, target resolution rate, target average response time, target reception time and target reception volume.

[0058] Among them, for example, satisfaction can be divided into five levels: "very dissatisfied, dissatisfied, average, satisfied, very satisfied", which are mapped to between 0 and 1, corresponding to [0 0.25 0.5 0.751] respectively; the resolution rate can be divided into two levels: "resolved, unresolved", which are mapped to between 0 and 1, corresponding to [0 1] respectively; the average response time and reception time are mapped to between 0 and 1 respectively. For example, the mapping formula can be expressed as:

[0059]

[0060] Among them, p represents the average response time or reception time obtained after mapping, and T represents the collected average response time and reception time.

[0061] Specifically, the reception volume can be used to represent the number of users that the online customer service can receive at the same time, and can also be used to represent the cumulative number of users received by the telephone customer service during the duty period. Exemplarily, the mapping formula of the average reply time and the reception time is used to map the reception volume to between 0 and 1 to obtain the mapped reception volume.

[0062] S130: Based on the target customer service indicator data, a target customer service matching the user to be tested is obtained from each preset customer service.

[0063] In a specific embodiment, based on each target customer service indicator data, a target customer service that matches the user to be tested is obtained from each preset customer service, including: for each preset customer service, based on each target indicator data in the target customer service indicator data corresponding to the preset customer service, determining the score to be tested corresponding to the preset customer service; and using the preset customer service with the highest score to be tested as the target customer service that matches the user to be tested.

[0064] Among them, illustratively, the score to be tested satisfies the formula:

[0065] ρ score =α satis ρ statis +α solve ρ solve +α response ρ response +α dur ρ dur +α satur ρ satur

[0066] Among them, ρ statis , solve , response , dur , satur They represent target satisfaction, target resolution rate, target average response time, target reception time and target reception volume, respectively. satis , α solve , α response , α dur , α satur Indicates the indicator weight corresponding to each target indicator data.

[0067] In another specific embodiment, a preset customer service whose score to be tested meets a preset score threshold is used as a target customer service matched with the user to be tested. Exemplarily, the preset score threshold may be 0.9, and the parameter value of the preset score threshold is not limited here, and can be customized according to actual needs.

[0068] In a specific embodiment, based on each target indicator data in the target customer service indicator data corresponding to the preset customer service, determining the score to be tested corresponding to the preset customer service includes: standardizing each target indicator data corresponding to the preset customer service to obtain at least one standardized target indicator data; and determining the score to be tested corresponding to the preset customer service based on each standardized customer service indicator data.

[0069] Since T≥1 in the mapping formula of average response time, reception time and reception volume, the average response time, reception time and reception volume after mapping are all greater than 0.5, while the satisfaction and resolution rate are less than 0.5. If the target indicator data are directly added together, the score to be tested will be biased towards the three indicators of average response time, reception time and reception volume. Considering that the variance of the five target indicator data is small, the embodiment of the present invention performs standardization processing on each target indicator data respectively, and performs weighted summation based on the standardized target indicator data to obtain the final score to be tested, which can further improve the accuracy of the score to be tested, and further improve the accuracy of the customer service allocation result.

[0070] The technical solution of this embodiment, in response to detecting a customer service allocation request of a tested user, inputs at least two types of feature data to be tested in at least one group of matched data combinations to be tested into a pre-trained target customer service indicator model, respectively, to obtain output target customer service indicator data corresponding to the preset customer services in each matched data combination to be tested, and based on each target customer service indicator data, obtains the target customer service matching the tested user from each preset customer service, wherein each feature data to be tested includes the tested user data corresponding to the tested user and the tested customer service data corresponding to the preset customer service. This embodiment constructs the correlation between the target customer service indicator data and the tested user data, solves the problem of mismatch between user consulting needs and the level of manual customer service, improves the accuracy of subsequent customer service allocation, and thereby improves the service quality and order facilitation rate of customer service.

[0071] Figure 3 This is a flow chart of another customer service allocation method provided by an embodiment of the present invention. This embodiment further refines the above-mentioned embodiment of "inputting each feature data to be tested in each matching data combination to be tested into a pre-trained target customer service indicator model to obtain output target customer service indicator data corresponding to the preset customer services in each matching data combination to be tested". Figure 3 As shown, the method includes:

[0072] S210: In response to detecting a customer service assignment request of a user to be tested, obtaining at least one set of matching data combinations to be tested.

[0073] S210 in this embodiment is similar to that in the above embodiment. Figure 1 The S110 shown corresponds to the same or similar steps, and will not be described in detail in this embodiment.

[0074] S220: For each matching data combination to be tested, input each feature data to be tested in the matching data combination to be tested into a pre-trained target customer service indicator model.

[0075] In this embodiment, the target customer service indicator model includes a feature encoding network and a feature decoding network.

[0076] S230. Perform feature coding processing on each input feature data to be tested through the feature coding network in the target customer service indicator model to obtain at least two coded features to be tested.

[0077] In a specific embodiment, the feature coding network includes a feature coding module, and each feature data to be tested shares a feature coding module, and the feature coding module is used to perform feature coding processing on each input feature data to be tested respectively, and output at least two coding features to be tested.

[0078] The advantage of this setting is that it can reduce the model size of the target customer service indicator model and reduce the memory usage of the target customer service indicator model.

[0079] In another specific embodiment, the feature coding network includes feature coding modules corresponding to each feature data to be tested, and the feature coding modules are used to perform feature coding processing on the input feature data to be tested and output the coding features to be tested.

[0080] The advantage of this setting is that it can realize parallel feature encoding processing of each input feature data to be tested, improve the prediction efficiency of the target customer service indicator model to meet the requirements of allocation efficiency in the customer service allocation scenario, and shorten the waiting time for manual customer service of the users to be tested.

[0081] Exemplarily, the feature encoding module may use an encoder in a Transformer model, which includes a multi-head attention layer, a fully connected layer, and two add & norm layers, wherein the add & norm layer includes a residual network and a layer normalization layer. The network architecture of the feature encoding module is not limited here.

[0082] In a specific embodiment, the feature encoding module adopts a STE model, wherein the STE model includes at least one encoder and an output layer in a Transformer model, wherein the exemplary output layer may be composed of multiple fully connected layers.

[0083] S240. Obtain the key data to be tested in each feature data to be tested through the feature decoding network in the target customer service indicator model, and output the target customer service indicator data corresponding to the preset customer service in the matching data combination to be tested based on the key data to be tested and each coding feature to be tested output by the feature encoding network.

[0084] In this embodiment, the key data to be tested represents the feature data pre-defined in each feature data to be tested. Specifically, the key data to be tested is customized in advance based on each feature data to be tested and the actual situation. For example, the key data to be tested includes but is not limited to whether it is a new user, whether it is a new customer service, customer service praise rate and average reception time, etc. The key data to be tested is not limited here, and can be customized according to actual needs.

[0085] In a specific embodiment, the feature decoding network includes a feature decoding module, and each target indicator data shares a feature decoding module. The feature decoding module is used to output the target customer service indicator data corresponding to the preset customer service based on the key data to be tested and the various coding features to be tested output by the feature encoding network.

[0086] The advantage of such a setting is that, when there are multiple target indicator data in the target customer service indicator data, the model size of the target customer service indicator model can be reduced, and the memory usage of the target customer service indicator model can be reduced.

[0087] In another specific embodiment, the feature decoding network includes feature decoding modules corresponding to each target indicator data, and the feature decoding module is used to output the target indicator data corresponding to the preset customer service in the matching data combination to be tested based on the key data to be tested and the coding features to be tested output by the feature encoding network.

[0088] Specifically, each feature decoding module outputs a target indicator data. The advantage of this setting is that it achieves the purpose of outputting the target customer service indicator data in parallel, improves the prediction efficiency of the target customer service indicator model, meets the requirements of allocation efficiency in the customer service allocation scenario, and shortens the waiting time for the tested user to see the manual customer service.

[0089] In a specific embodiment, the feature decoding module includes a multi-task weight unit, a convolution layer and an output layer. The multi-task weight unit is used to determine the key feature weight data based on the key data to be tested, and output the weight feature to be tested based on the key feature weight data and each coding feature to be tested output by the feature encoding network; the convolution layer is used to convolve the weight feature to be tested output by the multi-task weight unit to obtain the convolution weight feature; the output layer is used to output the target indicator data corresponding to the preset customer service in the matching data combination to be tested based on the convolution weight feature output by the convolution layer.

[0090] Specifically, the multi-task weight learning module (MWLM) includes a fully connected layer, a sigmoid function and a weighted concatenation layer, wherein the fully connected layer and the sigmoid function can convert each key feature in the key data to be tested into a feature weight, respectively, to obtain key feature weight data, and the weighted concatenation layer is used to multiply the key feature weight data and each coded feature to be tested output by the feature coding network bit by bit to obtain at least two weighted features to be tested, and concatenate each weighted feature to be tested to obtain the weighted feature to be tested. Among them, the feature weight can reflect the need to enhance or weaken the predicted customer service index data for each feature data to be tested.

[0091] Figure 4 A network architecture diagram of a multi-task weight unit provided by an embodiment of the present invention. Specifically, Figure 4 The large dotted box represents the multi-task weight unit, the small dotted box represents the weighted splicing layer in the multi-task weight unit, the black rectangle below represents the key feature data in the key data to be tested, the black rectangle on the left represents the 4 coding features to be tested output by the feature coding network, and the black rectangles of different grayscales in the small dotted box represent the weighted features to be tested obtained by bitwise multiplication of the 4 coding features to be tested based on the key feature weight data.

[0092] Since the target indicator data are not simply positively correlated, different target indicator data may restrict each other, thus affecting the training process of the target customer service indicator model. The multi-task weight unit can weight each feature data to be tested through adaptive weights according to the indicator characteristics of each target indicator data, which can effectively prevent model collapse.

[0093] Among them, exemplarily, the output layer in the feature decoding module includes a fully connected layer and a sigmoid function. The network architecture of the output layer is not limited here and can be customized according to actual needs.

[0094] Figure 5 A network architecture diagram of a specific example of a target customer service indicator model provided by an embodiment of the present invention. Specifically, Figure 5 The “STE model” in the figure represents the feature encoding module in the feature encoding network, and the “MWLM” represents the multi-task weight unit in the feature decoding module in the feature decoding network. Figure 5 In the target customer service indicator model shown, the feature encoding network includes feature encoding modules corresponding to each feature data to be measured, and the feature decoding network includes feature decoding modules corresponding to each target indicator data.

[0095] S250: Based on the target customer service indicator data, a target customer service matching the user to be tested is obtained from each preset customer service.

[0096] S250 in this embodiment is different from S250 in the above embodiment. Figure 1 The S130 shown corresponds to the same or similar steps, and will not be described in detail in this embodiment.

[0097] The technical solution of this embodiment, for each matching data combination to be tested, inputs each feature data to be tested in the matching data combination to be tested into a pre-trained target customer service indicator model, performs feature encoding processing on each input feature data to be tested through the feature encoding network in the target customer service indicator model, obtains at least two coded features to be tested, obtains key data to be tested in each feature data to be tested through the feature decoding network in the target customer service indicator model, and outputs target customer service indicator data corresponding to the preset customer service in the matching data combination to be tested based on the key data to be tested and each coded feature to be tested output by the feature encoding network, so that the target customer service indicator model can perform weighted control on each input feature data to be tested based on the key data to be tested, thereby solving the problem of inaccurate target customer indicator data output by the traditional target customer indicator model, further improving the accuracy of subsequent customer service allocation, and further improving the service quality and order completion rate of customer service.

[0098] Figure 6 This is a flowchart of a method for training a customer service indicator model provided by an embodiment of the present invention. This embodiment is applicable to the case of training a neural network model for predicting customer service-related indicator scores. The method can be executed by a training device for a customer service indicator model. The training device for the customer service indicator model can be implemented in the form of hardware and / or software. The training device for the customer service indicator model can be configured in a terminal device. Figure 6 As shown, the method includes:

[0099] S310: Obtain at least one set of training sample data combinations.

[0100] In this embodiment, the training sample data combination includes at least two types of training feature data, each of which includes training user data corresponding to the training user and training customer service data corresponding to the training customer service. In a specific embodiment, the training sample data combination also includes training scene data and / or training text data.

[0101] The "training sample data combination" in this embodiment corresponds to the same or similar to the "matching data combination to be tested" in the above embodiment, and will not be described in detail in this embodiment.

[0102] S320: Input each training feature data in each training sample data combination into an untrained initial customer service indicator model, and obtain output predicted customer service indicator data corresponding to the training customer service in each training sample data combination.

[0103] In a specific embodiment, the initial customer service indicator model includes a feature encoding network and a feature decoding network, and each training feature data in each training sample data combination is respectively input into the untrained initial customer service indicator model to obtain output predicted customer service indicator data corresponding to the training customer service in each training sample data combination, including: for each training sample data combination, each training feature data in the training sample data combination is input into the untrained initial customer service indicator model; through the feature encoding network, each input training feature data is respectively feature encoded to obtain at least two training coding features; through the feature decoding network, the training key data in each training feature data is obtained, and based on the training key data and each training coding feature output by the feature encoding network, the predicted customer service indicator data corresponding to the training customer service in the training sample data combination is output; wherein the training key data represents the pre-defined feature data in each training feature data.

[0104] The "predicted customer service indicator data" in this embodiment corresponds to the same or similar to the "target customer service indicator data" in the above embodiment, and will not be described in detail in this embodiment.

[0105] In a specific embodiment, the feature encoding network includes feature encoding modules corresponding to each training feature data, and the feature decoding network includes feature decoding modules corresponding to each prediction index data.

[0106] In a specific embodiment, the feature decoding module includes a multi-task weight unit, a convolution layer and an output layer. The multi-task weight unit is used to determine the key feature weight data based on the training key data, and output the training weight features based on the key feature weight data and the training coding features output by the feature encoding network; the convolution layer is used to convolve the training weight features output by the multi-task weight unit to obtain the convolution weight features; the output layer is used to output the prediction index data corresponding to the preset customer service in the training sample data combination based on the convolution weight features output by the convolution layer.

[0107] S330: Determine a training loss function based on each predicted customer service indicator data and the standard customer service indicator data corresponding to each training sample data combination.

[0108] In a specific embodiment, each predicted customer service indicator data includes at least one of predicted indicator data of predicted satisfaction, predicted resolution rate, predicted average response time, predicted reception time, and predicted reception volume. Correspondingly, the standard customer service indicator data includes at least one of standard indicator data of standard satisfaction, standard resolution rate, standard average response time, standard reception time, and standard reception volume.

[0109] Figure 7 An architectural diagram of a specific instance of a training sample data combination and standard customer service indicator data provided by an embodiment of the present invention. Specifically, feature extraction is performed on historical user consultation-customer service reception data to obtain four types of training feature data, namely, training user data, training customer service data, training scenario data, and training text data. Label extraction is performed on historical user consultation-customer service reception data to obtain five types of standard indicator data, namely, standard satisfaction and standard resolution rate related to user experience, standard average response time and standard reception time related to service efficiency, and standard reception volume related to reception fairness.

[0110] Specifically, based on each prediction indicator data and the standard indicator data corresponding to each prediction indicator data, loss functions are constructed respectively, based on each loss function, a training loss function is determined, and based on the training loss function, the model parameters of the initial customer service indicator model are adjusted.

[0111] Specifically, the function types corresponding to the loss functions can be the same or different. Exemplarily, the function types of the loss function include but are not limited to the square loss function, logarithmic loss function, exponential loss function, mean square error loss function, logistic regression loss function, Huber loss function, cross entropy loss function and Kullback-Leibler divergence loss function, etc. The function type of the loss function is not limited here, and can be customized according to actual needs.

[0112] In a specific embodiment, the function types corresponding to the loss functions are all cross entropy loss functions.

[0113] Among them, exemplary, training loss function Satisfy the formula:

[0114]

[0115] in, represents the loss function determined based on the predicted satisfaction, predicted resolution rate, predicted average response time, predicted reception time, and predicted reception volume, respectively. satis , α solve , α response , α dur , α satur Represents the function weights corresponding to each pre-defined loss function.

[0116] S340. Based on the training loss function, adjust the model parameters of the initial customer service indicator model until the training loss function converges, and use the initial customer service indicator model as the target customer service indicator model after training.

[0117] In a specific embodiment, the model parameters of the initial customer service index model are adjusted based on the training loss function, including: adjusting the model parameters of the feature encoding network in the initial customer service index model based on the training loss function; adjusting the weight parameters corresponding to each training feature data in the feature decoding network in the initial customer service index model based on the training loss function.

[0118] Since the feature decoding network contains a multi-task weight unit, the multi-task weight unit can adaptively adjust its weight coefficient according to the indicator characteristics of each prediction indicator data during the training process of the customer service indicator model to perform weighted control on each feature data to be tested, which can effectively prevent model collapse.

[0119] The technical solution of this embodiment is to input each training feature data in at least one group of acquired training sample data combinations into an untrained initial customer service indicator model, obtain output predicted customer service indicator data corresponding to the training customer service in each training sample data combination, determine the training loss function based on each predicted customer service indicator data and the standard customer service indicator data corresponding to each training sample data combination, and adjust the model parameters of the initial customer service indicator model based on the training loss function until the training loss function converges. The initial customer service indicator model is used as the trained target customer service indicator model, so that the customer service indicator model trained by the embodiment of the present invention can perform weighted control on each input feature data to be tested based on the key data to be tested, thereby solving the problem of inaccurate target customer indicator data output by the traditional target customer indicator model, improving the accuracy of subsequent customer service allocation, and thereby improving the service quality and order facilitation rate of customer service.

[0120] Figure 8 This is a flow chart of another customer service indicator model training method provided by an embodiment of the present invention. This embodiment further refines the "feature encoding network" in the above embodiment. Figure 8 As shown, the method includes:

[0121] S410: Obtain a pre-trained target text encoding module corresponding to the training text data, and construct an initial customer service indicator model based on the target text encoding module.

[0122] In a specific embodiment, a pre-trained target text encoding module corresponding to training text data is obtained, including: masking the obtained pre-trained text data to obtain masked text data; inputting the masked text data into an initial text encoding module that has not been pre-trained to obtain output predicted text data; and adjusting the model parameters of the initial text encoding module based on the predicted text data and the pre-trained text data to obtain the pre-trained target text encoding module.

[0123] Exemplarily, the mask processing may be to replace at least one word in a sentence in the pre-trained text data with "mask" or other random values ​​to obtain masked text data.

[0124] In a specific embodiment, the initial text encoding module adopts a STE model, wherein the STE model includes at least one encoder and an output layer in a Transformer model. The exemplary output layer may be composed of multiple fully connected layers.

[0125] In a specific embodiment, based on the predicted text data and the pre-trained text data, the model parameters of the initial text encoding module are adjusted to obtain a pre-trained target text encoding module, including: constructing a target loss function based on the predicted text data and the pre-trained text data, and adjusting the model parameters of the initial text encoding module based on the target loss function until the target loss function converges, and then using the initial text encoding module as the pre-trained target text encoding module.

[0126] Exemplarily, the function type of the target loss function includes, but is not limited to, a square loss function, a logarithmic loss function, an exponential loss function, a mean square error loss function, a logistic regression loss function, a Huber loss function, a cross entropy loss function, or a Kullback-Leibler divergence loss function, etc. The function type of the target loss function is not limited here, and can be customized according to actual needs.

[0127] In another specific embodiment, the target text encoding module is pre-trained using a knowledge transfer method, and the target text encoding module includes at least one encoder and an output layer connected in series. Accordingly, based on the predicted text data and the pre-trained text data, the model parameters of the initial text encoding module are adjusted to obtain the pre-trained target text encoding module, including: inputting the mask text data into the trained reference text model to obtain the output reference text data; based on the number of encoders in the target text encoding module, obtaining the reference text features respectively output by at least one encoder in the reference text model; based on the reference text data, each reference text feature, the predicted text data and the pre-trained text data, constructing a target loss function; based on the target loss function, adjusting the model parameters of the initial text encoding module to obtain the pre-trained target text encoding module.

[0128] Specifically, the reference text model can be used to represent a feature extraction model with a large model volume and accurate prediction results. In a specific embodiment, the reference text model is a BERT model. The network architecture of the BERT model is generally composed of 24 encoders of the transformer network. The trained BERT model has a strong feature extraction capability and can use the multi-head self-attention mechanism to mine the grammatical and semantic relationships within the training text.

[0129] Specifically, the number of encoders in the reference text model is greater than the number of encoders in the target text encoding module.

[0130] In a specific embodiment, a target loss function is constructed based on reference text data, each reference text feature, predicted text data and pre-trained text data, including: obtaining intermediate text features respectively output by each encoder in the target text encoding module; constructing a first loss function based on the predicted text data and the pre-trained text data, and constructing a second loss function based on the predicted text data and the reference text data; constructing at least one third loss function based on each reference text feature and each intermediate text feature; constructing a target loss function based on the first loss function, the second loss function and each third loss function.

[0131] Specifically, the function types of the first loss function, the second loss function and the third loss function may be the same or different.

[0132] In a specific embodiment, the function type of the first loss function is a cross entropy loss function. Satisfy the formula:

[0133]

[0134] Among them, n represents the total number of words replaced in the mask text data, y i represents the probability distribution of the pre-trained text data relative to the i-th replacement word, Represents the probability distribution of the predicted text data relative to the i-th replacement word.

[0135] In a specific embodiment, the function type of the second loss function is a Kullback-Leibler divergence loss function. Satisfy the formula:

[0136]

[0137] in, represents the probability distribution of the reference text data relative to the i-th replacement word, Represents the probability distribution of the BERT model output after optimization of the temperature coefficient T, Represents the probability distribution of the STE model output after the temperature coefficient T is optimized.

[0138] Wherein, illustratively, the temperature coefficient T is optimized to satisfy the formula:

[0139]

[0140] in, represents the output vector corresponding to the i-th replacement word output by the BERT model, and k represents the total number of replaced words in the masked text data.

[0141] Wherein, illustratively, T = 2. The advantage of adding the temperature coefficient T is that the STE model does not rigidly learn the knowledge of the BERT model, but has flexibility, thereby improving the generalization ability of the STE model.

[0142] In a specific embodiment, the function type of the third loss function is a mean square error loss function. Exemplarily, the third loss function satisfies the formula:

[0143]

[0144] Specifically, the reference text features and the intermediate text features are matrices, Represents the xth row and yth column element of the intermediate text feature output by the lth encoder in the STE model, represents the lth encoder in the BERT model B The x-th row and y-th column element of the reference text feature output by the encoder, m represents the number of encoders in the STE model.

[0145] Wherein, for example, when m=3, l and l B The corresponding relationship can be [(1, 8), (2, 16), (3, 20)]. There is no limitation on the corresponding relationship between the encoders in the STE model and the encoders in the BERT model.

[0146] Among them, exemplary, the objective loss function satisfies the formula:

[0147]

[0148] Among them, α CE , α KL and α MSE Represents the harmonic weights corresponding to the first loss function, the second loss function, and the third loss function respectively.

[0149] Fig. 9 A schematic diagram of a training method for a target text encoding module provided by an embodiment of the present invention. Specifically, Fig. 9 The dotted box on the left represents the BERT model. Specifically, the BERT model contains 24 encoders and an output layer with fixed parameters. Fig. 9 The dashed box on the right represents the STE model, which contains three encoders and an output layer whose parameters are to be adjusted. The first loss function is determined based on the pre-trained text data and the predicted text data output by the STE model, the second loss function is constructed based on the predicted text data output by the STE model and the reference text data output by the BERT model, and the third loss function is constructed based on the intermediate text features output by the three encoders in the STE model and the reference text features output by the corresponding three encoders in the BERT model.

[0150] The advantage of this setting is that compared with a single training process, the STE model obtained by pre-training the reference text model through transfer learning not only has the prediction effect of the reference text model, but also has a smaller model size and faster prediction speed than the reference text model.

[0151] Specifically, when constructing the initial customer service indicator model, the feature encoding module corresponding to the training text data in the feature encoding network is the target text encoding module obtained through the above-mentioned pre-training process, and the feature encoding modules corresponding to other training feature data are feature encoding modules with randomly initialized parameters.

[0152] S420: Obtain at least one set of training sample data combinations.

[0153] S430, input each training feature data in each training sample data combination into an untrained initial customer service indicator model, and obtain output predicted customer service indicator data corresponding to the training customer service in each training sample data combination.

[0154] S440: Determine a training loss function based on each predicted customer service indicator data and the standard customer service indicator data corresponding to each training sample data combination.

[0155] S450. Based on the training loss function, adjust the model parameters of the initial customer service indicator model until the training loss function converges, and use the initial customer service indicator model as the target customer service indicator model after training.

[0156] S420-S450 in this embodiment are similar to those in the above embodiment. Figure 6 The S310 - S340 shown correspond to the same or similar ones, and are not described in detail in this embodiment.

[0157] The technical solution of this embodiment solves the problem of low training efficiency of the initial customer service indicator model by constructing an initial customer service indicator model based on a pre-trained target text encoding module. During the training process of the customer service indicator model, it is only necessary to fine-tune the target text encoding module in the initial customer service indicator model. This improves the training efficiency of the customer service indicator model while ensuring the prediction effect of the customer service indicator model.

[0158] The following is an embodiment of a customer service allocation device provided in an embodiment of the present invention. The device and the customer service allocation method of the above embodiment belong to the same inventive concept. For details not fully described in the embodiment of the customer service allocation device, please refer to the content about the customer service allocation method in the above embodiment.

[0159] Fig.10This is a schematic diagram of the structure of a customer service distribution device provided by an embodiment of the present invention. Fig.10 As shown, the device includes: a matching data combination acquisition module 510 to be tested, a target customer service indicator data output module 520 and a target customer service matching module 530.

[0160] The module 510 for acquiring the matching data combination to be tested is used to acquire at least one set of matching data combination to be tested in response to detecting a customer service allocation request of the user to be tested; wherein the matching data combination to be tested includes at least two types of feature data to be tested, each feature data to be tested includes user data to be tested corresponding to the user to be tested and customer service data to be tested corresponding to the preset customer service;

[0161] The target customer service indicator data output module 520 is used to input each feature data to be tested in each matching data combination to the pre-trained target customer service indicator model, and obtain the output target customer service indicator data corresponding to the preset customer service in each matching data combination to be tested;

[0162] The target customer service matching module 530 is used to obtain a target customer service that matches the user to be tested from each preset customer service based on the target customer service indicator data.

[0163] The technical solution of this embodiment solves the problem of mismatch between user consultation needs and manual customer service levels by building a correlation between target customer service indicator data and the user data to be tested, improves the accuracy of subsequent customer service allocation, and thereby improves the service quality and order completion rate of customer service.

[0164] In a specific embodiment, the target customer service indicator data output module 520 is specifically used to:

[0165] For each matching data combination to be tested, each feature data to be tested in the matching data combination to be tested is input into a pre-trained target customer service indicator model; wherein the target customer service indicator model includes a feature encoding network and a feature decoding network;

[0166] Through the feature coding network, feature coding processing is performed on each input feature data to be tested, and at least two coding features to be tested are obtained;

[0167] Through the feature decoding network, the key data to be tested in each feature data to be tested is obtained, and based on the key data to be tested and each coding feature to be tested output by the feature encoding network, the target customer service indicator data corresponding to the preset customer service in the combination of matching data to be tested is output;

[0168] The key data to be tested represents the predefined characteristic data in each characteristic data to be tested.

[0169] In a specific embodiment, the target customer service indicator data includes at least one target indicator data of target satisfaction, target resolution rate, target average response time, target reception time and target reception volume;

[0170] The feature decoding network includes feature decoding modules corresponding to each target indicator data. The feature decoding module is used to output the target indicator data corresponding to the preset customer service in the matching data combination to be tested based on the key data to be tested and the coding features to be tested output by the feature encoding network.

[0171] In a specific embodiment, the feature decoding module includes a multi-task weight unit, a convolution layer and an output layer. The multi-task weight unit is used to determine the key feature weight data based on the key data to be tested, and output the weight feature to be tested based on the key feature weight data and each coding feature to be tested output by the feature encoding network; the convolution layer is used to convolve the weight feature to be tested output by the multi-task weight unit to obtain the convolution weight feature; the output layer is used to output the target indicator data corresponding to the preset customer service in the matching data combination to be tested based on the convolution weight feature output by the convolution layer.

[0172] In a specific embodiment, the matching data combination to be tested also includes scene data to be tested and / or text data to be tested, and the feature coding network includes feature coding modules corresponding to each feature data to be tested, and the feature coding module is used to perform feature coding processing on the input feature data to be tested to obtain the coding feature to be tested.

[0173] In a specific embodiment, the target customer service matching module 530 includes:

[0174] A score determination unit for determining a score to be tested, for each preset customer service, based on each target indicator data in the target customer service indicator data corresponding to the preset customer service, to determine a score to be tested corresponding to the preset customer service;

[0175] The target customer service matching unit is used to use the preset customer service with the highest score to be tested as the target customer service to be matched with the user to be tested.

[0176] In a specific embodiment, the unit for determining the score to be tested is specifically used to:

[0177] Standardize each target indicator data corresponding to the preset customer service to obtain at least one standardized target indicator data;

[0178] Based on the standardized customer service indicator data, determine the test score corresponding to the preset customer service.

[0179] The customer service allocation device provided in the embodiment of the present invention can execute the customer service allocation method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0180] The following is an embodiment of a training device for a customer service indicator model provided in an embodiment of the present invention. The device and the training method for the customer service indicator model in the above-mentioned embodiment belong to the same inventive concept. For details not fully described in the embodiment of the training device for the customer service indicator model, reference can be made to the contents of the training method for the customer service indicator model in the above-mentioned embodiment.

[0181] Fig.11 A schematic diagram of a training device for a customer service indicator model provided by an embodiment of the present invention. Fig.11 As shown, the device includes: a training sample data combination acquisition module 610, a predicted customer service index data output module 620, a training loss function determination module 630 and a target customer service index model determination module 640.

[0182] The training sample data combination acquisition module 610 is used to acquire at least one set of training sample data combination; wherein the training sample data combination includes at least two types of training feature data, each of which includes training user data corresponding to the training user and training customer service data corresponding to the training customer service;

[0183] The predicted customer service indicator data output module 620 is used to input each training feature data in each training sample data combination into the untrained initial customer service indicator model, and obtain the output predicted customer service indicator data corresponding to the training customer service in each training sample data combination;

[0184] A training loss function determination module 630, for determining a training loss function based on each predicted customer service indicator data and the standard customer service indicator data corresponding to each training sample data combination;

[0185] The target customer service indicator model determination module 640 is used to adjust the model parameters of the initial customer service indicator model based on the training loss function until the training loss function converges, and then use the initial customer service indicator model as the trained target customer service indicator model.

[0186] The customer service indicator model trained by the technical solution of this embodiment can perform weighted control on the input feature data to be tested based on the key data to be tested, thereby solving the problem of inaccurate target customer indicator data output by the traditional target customer indicator model, improving the accuracy of subsequent customer service allocation, and thereby improving the service quality and order completion rate of customer service.

[0187] In a specific embodiment, the initial customer service indicator model includes a feature encoding network and a feature decoding network. The feature encoding network includes feature encoding modules corresponding to each training feature data. When the training sample data combination also includes training text data, the feature encoding module corresponding to the training text data in the feature encoding network is a pre-trained target text encoding module, and the target text encoding module is pre-trained using a knowledge transfer method.

[0188] In a specific embodiment, the target customer service indicator model determination module 640 is specifically used to:

[0189] Based on the training loss function, adjust the model parameters of the feature encoding network in the initial customer service indicator model;

[0190] Based on the training loss function, the weight parameters corresponding to each training feature data in the feature decoding network in the initial customer service indicator model are adjusted.

[0191] The training device for the customer service indicator model provided in the embodiment of the present invention can execute the training method for the customer service indicator model provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0192] Fig.12 A schematic diagram of the structure of an electronic device provided for one embodiment of the present invention. The electronic device 10 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0193] like Fig.12As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0194] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0195] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The processor 11 executes the various methods and processes described above, such as the training method of the customer service indicator model provided in the above embodiments, and / or the customer service allocation method.

[0196] In some embodiments, the training method of the customer service indicator model provided in the above embodiments, and / or the customer service allocation method may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the training method of the customer service indicator model and / or the customer service allocation method described above may be executed. Alternatively, in other embodiments, the processor 11 may be configured to execute the training method of the customer service indicator model and / or the customer service allocation method in any other appropriate manner (e.g., by means of firmware).

[0197] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0198] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0199] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0200] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0201] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0202] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0203] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0204] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A customer service allocation method, characterized in that: include: In response to detecting a customer service allocation request of a user to be tested, obtaining at least one set of matching data combinations to be tested; wherein the matching data combinations to be tested include at least two feature data to be tested, each of which includes user data to be tested corresponding to the user to be tested and customer service data to be tested corresponding to a preset customer service; Input each feature data to be tested in each of the matching data combinations to be tested into a pre-trained target customer service indicator model, and obtain output target customer service indicator data corresponding to the preset customer services in each of the matching data combinations to be tested; Based on the target customer service indicator data, a target customer service matching the user to be tested is obtained from the preset customer services.

2. The method according to claim 1, characterized in that Inputting each feature data to be tested in each of the matching data combinations to be tested into a pre-trained target customer service indicator model, respectively, to obtain output target customer service indicator data corresponding to the preset customer service in each of the matching data combinations to be tested, including: For each matching data combination to be tested, each feature data to be tested in the matching data combination to be tested is input into a pre-trained target customer service indicator model; wherein the target customer service indicator model includes a feature encoding network and a feature decoding network; Through the feature coding network, feature coding processing is performed on each of the input feature data to be tested to obtain at least two coded features to be tested; Obtaining key data to be tested in each of the feature data to be tested through the feature decoding network, and outputting target customer service indicator data corresponding to the preset customer service in the combination of matching data to be tested based on the key data to be tested and each of the coding features to be tested output by the feature encoding network; The key data to be tested represents the predefined characteristic data in each characteristic data to be tested.

3. The method according to claim 2, characterized in that The target customer service indicator data includes at least one target indicator data of target satisfaction, target resolution rate, target average response time, target reception time and target reception volume; The feature decoding network includes feature decoding modules corresponding to each of the target indicator data, and the feature decoding modules are used to output the target indicator data corresponding to the preset customer service in the matching data combination to be tested based on the key data to be tested and each of the coding features to be tested output by the feature encoding network.

4. The method according to claim 3, characterized in that The feature decoding module includes a multi-task weight unit, a convolution layer and an output layer. The multi-task weight unit is used to determine the key feature weight data based on the key data to be tested, and output the weight feature to be tested based on the key feature weight data and each of the coded features to be tested output by the feature encoding network; the convolution layer is used to perform convolution processing on the weight feature to be tested output by the multi-task weight unit to obtain the convolution weight feature; the output layer is used to output the target indicator data corresponding to the preset customer service in the matching data combination to be tested based on the convolution weight feature output by the convolution layer.

5. The method according to claim 2, characterized in that: The matching data combination to be tested also includes scene data to be tested and / or text data to be tested, and the feature coding network includes feature coding modules corresponding to each of the feature data to be tested, and the feature coding module is used to perform feature coding processing on the input feature data to be tested to obtain the coding features to be tested.

6. The method according to claim 3, characterized in that The step of acquiring a target customer service matching the user to be tested from each of the preset customer services based on each of the target customer service indicator data includes: For each preset customer service, based on each target indicator data in the target customer service indicator data corresponding to the preset customer service, determine the score to be tested corresponding to the preset customer service; The preset customer service with the highest score to be tested is used as the target customer service to match the user to be tested.

7. The method according to claim 6, characterized in that The step of determining the score to be tested corresponding to the preset customer service based on each target indicator data in the target customer service indicator data corresponding to the preset customer service comprises: Standardizing each of the target indicator data corresponding to the preset customer service to obtain at least one standardized target indicator data; Based on each standardized customer service indicator data, determine the score to be tested corresponding to the preset customer service.

8. A method for training a customer service indicator model, characterized in that: include: Acquire at least one set of training sample data combination; wherein the training sample data combination includes at least two types of training feature data, each of the training feature data includes training user data corresponding to the training user and training customer service data corresponding to the training customer service; Input each training feature data in each of the training sample data combinations into an untrained initial customer service indicator model to obtain output predicted customer service indicator data corresponding to the training customer service in each of the training sample data combinations; Determine a training loss function based on each of the predicted customer service indicator data and the standard customer service indicator data corresponding to each of the training sample data combinations; Based on the training loss function, the model parameters of the initial customer service indicator model are adjusted until the training loss function converges, and the initial customer service indicator model is used as the target customer service indicator model after training.

9. The method according to claim 8, characterized in that The initial customer service indicator model includes a feature encoding network and a feature decoding network. The feature encoding network contains feature encoding modules corresponding to each of the training feature data. When the training sample data combination also includes training text data, the feature encoding module corresponding to the training text data in the feature encoding network is a pre-trained target text encoding module, and the target text encoding module is pre-trained using a knowledge transfer method.

10. The method according to claim 9, characterized in that The adjusting the model parameters of the initial customer service indicator model based on the training loss function includes: Based on the training loss function, adjusting the model parameters of the feature encoding network in the initial customer service indicator model; Based on the training loss function, the weight parameters corresponding to each of the training feature data in the feature decoding network in the initial customer service indicator model are adjusted.

11. A customer service allocation device, characterized in that: include: The module for acquiring the combination of matching data to be tested is used to acquire at least one group of combination of matching data to be tested in response to detecting a customer service allocation request of the user to be tested; wherein the combination of matching data to be tested includes at least two types of feature data to be tested, each of which includes user data to be tested corresponding to the user to be tested and customer service data to be tested corresponding to a preset customer service; A target customer service indicator data output module is used to input each feature data to be tested in each matching data combination to a pre-trained target customer service indicator model, and obtain output target customer service indicator data corresponding to the preset customer service in each matching data combination to be tested; The target customer service matching module is used to obtain the target customer service that matches the user to be tested from each of the preset customer services based on the target customer service indicator data.

12. A training device for a customer service indicator model, characterized in that: include: A training sample data combination acquisition module is used to acquire at least one set of training sample data combinations; wherein the training sample data combination includes at least two types of training feature data, each of which includes training user data corresponding to a training user and training customer service data corresponding to a training customer service; A predicted customer service indicator data output module is used to input each training feature data in each of the training sample data combinations into an untrained initial customer service indicator model to obtain output predicted customer service indicator data corresponding to the training customer service in each of the training sample data combinations; A training loss function determination module, used to determine a training loss function based on each of the predicted customer service indicator data and the standard customer service indicator data corresponding to each of the training sample data combinations; The target customer service indicator model determination module is used to adjust the model parameters of the initial customer service indicator model based on the training loss function until the training loss function converges, and then use the initial customer service indicator model as the trained target customer service indicator model.

13. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the customer service allocation method described in any one of claims 1-7, and / or can execute the training method of the customer service indicator model described in any one of claims 8-10.

14. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, which are used to enable a processor to implement the customer service allocation method described in any one of claims 1-7 when executed, and / or to be able to execute the training method of the customer service indicator model described in any one of claims 8-10.

Images