AI digital employee training method based on deep learning
By constructing a deep learning-based AI digital employee training method, combining employee work history data of target application scenarios and associated application scenarios, extracting and registering Q&A features, the problem that traditional training methods cannot cope with multiple needs is solved, and AI digital is improved Employees’ coping skills and customer experience.
Patent Information
- Application Number
- CN202410929646.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-11
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-07-11
AI Technical Summary
Traditional AI digital employee training methods cannot effectively respond to customer needs, resulting in insufficient coping skills and poor customer experience.
By determining the target application scenarios and associated application scenarios of AI digital employees, retrieve corresponding employee work history data, perform feature extraction and registration, build training data sets, and use deep learning models to train AI digital employees to ensure that they are applicable Responsiveness to target application scenarios.
The coping ability of AI digital employees is improved to match the target scenarios, so that they can better respond to customer needs and improve customer experience.
Smart Images

Figure CN118586923B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of information technology, and in particular to an AI digital employee training method based on deep learning. Background Art
[0002] The digital economy is developing rapidly in today's world, and the labor model is also evolving rapidly. Through technological empowerment, traditional labor is combined with digital labor to establish "digital employees", freeing traditional labor from tedious procedural business and focusing on more valuable and creative work.
[0003] In order to make digital employees have strong enough response capabilities, it is necessary to use the measured data of specific scenarios to build a training set and use this training set to train digital employees. However, customer needs may be multifaceted. In addition to the response capabilities acquired by digital employees after training, customers may also consult about other areas. Digital employees trained by the above traditional training methods cannot respond, resulting in poor customer experience. Summary of the invention
[0004] In this regard, the present invention provides an AI digital employee training method, system, electronic device, computer storage medium and computer program product based on deep learning to solve the above-mentioned technical problems.
[0005] The present invention discloses an AI digital employee training method based on deep learning, the method comprising the following steps:
[0006] Determine a target application scenario of the AI digital employee and an associated application scenario associated with the target application scenario, and retrieve the first employee work history data corresponding to the target application scenario and the second employee work history data corresponding to the associated application scenario;
[0007] Performing feature extraction on the first employee's work history data and the second employee's work history data respectively to obtain a first question-and-answer feature set and a second question-and-answer feature set;
[0008] Using a deep learning model to align the elements in the first question-answer feature set and the second question-answer feature set, obtaining a plurality of question-answer feature pairs, each of which includes a first question-answer feature and a corresponding second question-answer feature, and forming each of the question-answer feature pairs into a training data set; wherein the first question-answer feature is associated with the first question-answer feature set, and the second question-answer feature is associated with the second question-answer feature set;
[0009] The AI digital employee is trained using the training data set so that the AI digital employee has a response capability suitable for the target application scenario.
[0010] Optionally, determining an associated application scenario associated with the target application scenario includes:
[0011] Retrieving each failure response record data corresponding to the target application scenario and performing semantic understanding on the data to obtain the scenario corresponding to each failure response record data;
[0012] Calculating the matching degree between each of the scenarios and the target application scenario, obtaining a number of target scenarios with matching degrees lower than a matching threshold, and sorting the target scenarios from large to small according to the matching degrees;
[0013] The response busyness of the target application scenario is determined, a specified number of the target scenarios ranked top are filtered out according to the response busyness, and the filtered target scenarios are determined as associated application scenarios.
[0014] Optionally, the extracting features of the first employee's work history data and the second employee's work history data to obtain a first question-and-answer feature set and a second question-and-answer feature set respectively includes:
[0015] Extracting features from the first employee's work history data to obtain the first question-and-answer feature set;
[0016] Extracting features from the second employee's work history data to obtain a number of initial question-and-answer features, and converting each of the initial question-and-answer features into natural language to obtain an initial question-and-answer dialogue;
[0017] A BERT model is used to perform semantic understanding on each of the initial question-and-answer dialogues to obtain each of the intermediate question-and-answer dialogues, a clustering algorithm is used to cluster each of the intermediate question-and-answer features into a number of target question-and-answer dialogues, and feature extraction is performed on each of the target question-and-answer dialogues to obtain a number of target question-and-answer features, each of which constitutes the second question-and-answer feature set.
[0018] Optionally, the using a deep learning model to align elements in the first question-answer feature set and the second question-answer feature set to obtain a plurality of groups of question-answer feature pairs includes:
[0019] The deep learning model groups the first question and answer features in the first question and answer feature set and the second question and answer features in the second question and answer feature set according to question and answer types to obtain a plurality of first groups and a plurality of second groups;
[0020] Obtaining a number of historical question-and-answer records corresponding to each of the failed response record data, calculating the average number of records of the historical question-and-answer records corresponding to all of the failed response record data, and determining the average number of records as the registration number; wherein the historical question-and-answer record is earlier in time than the failed response record data and belongs to a single question-and-answer session of the same customer;
[0021] A number of the registered number of the second groups are randomly selected to be associated with a certain first group to obtain a number of question-answer feature pairs.
[0022] Optionally, the deep learning model groups the first question and answer features in the first question and answer feature set and the second question and answer features in the second question and answer feature set according to question and answer types, including:
[0023] The deep learning model retrieves an operation specification file corresponding to the target application scenario, and determines and obtains a plurality of question and answer types according to the operation specification file;
[0024] The first question and answer features in the first question and answer feature set and the second question and answer features in the second question and answer feature set are grouped according to the question and answer type.
[0025] Optionally, the determining of a plurality of question and answer types according to the operation specification file includes:
[0026] The deep learning model parses the directory part in the operation specification file, determines the configuration quantity of the secondary blocks corresponding to each primary block, and calculates the average configuration quantity of each configuration quantity;
[0027] If the average configuration quantity exceeds a threshold, the deep learning model is set to determine and derive a number of the question and answer types according to the operation specification file; otherwise, the deep learning model is set not to determine and derive a number of the question and answer types according to the operation specification file.
[0028] The present invention also provides an AI digital employee training system based on deep learning, the system comprising a retrieval module, a feature extraction module, a training set building module, and a training module;
[0029] The retrieval module is used to determine the target application scenario of the AI digital employee and the associated application scenario associated with the target application scenario, and retrieve the first employee work history data corresponding to the target application scenario and the second employee work history data corresponding to the associated application scenario;
[0030] The feature extraction module is used to extract features from the first employee's work history data and the second employee's work history data respectively to obtain a first question-answer feature set and a second question-answer feature set;
[0031] The training set building module is used to use a deep learning model to align the elements in the first question-answer feature set and the second question-answer feature set to obtain a plurality of question-answer feature pairs, each of which contains a first question-answer feature and a corresponding second question-answer feature, and to build each of the question-answer feature pairs into a training data set; wherein the first question-answer feature is associated with the first question-answer feature set, and the second question-answer feature is associated with the second question-answer feature set;
[0032] The training module is used to train the AI digital employee using the training data set so that the AI digital employee has a response capability suitable for the target application scenario.
[0033] The present invention also provides an electronic device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program implements the method described in any of the preceding items when executed by the processor.
[0034] The present invention also provides a computer storage medium storing a computer program executable by a processor to implement any of the methods described above.
[0035] The present invention also provides a computer program product, which includes a computer program that can be executed by a processor to implement any of the methods described above.
[0036] The beneficial effects of the present invention are:
[0037] On the one hand, the present invention constructs a training data set based on the correlation between the answers provided by real employees to deal with specific problems in the scene, which can better reflect that the response ability acquired by the trained AI digital employees is more matched with the corresponding scene. On the other hand, the present invention not only selects the work data of real employees in the specific target application scenarios of AI digital employees, but also further expands to the work data of real employees in related application scenarios that have a certain correlation with the target application scenarios, and combines the data of the two to extract the correlation between the answers provided by real employees to deal with specific problems in the scene, that is, the question-answering features, so that the trained AI digital employees have stronger response capabilities and can bring a better experience to customers. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.
[0039] Figure 1 It is a flowchart of an AI digital employee training method based on deep learning disclosed in an embodiment of the present invention;
[0040] Figure 2 It is a structural diagram of an AI digital employee training system based on deep learning disclosed in an embodiment of the present invention. DETAILED DESCRIPTION
[0041] The following is a description of the implementation of the present application by specific specific embodiments. People familiar with the technology can easily understand other advantages and effects of the present application from the contents disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.
[0042] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.
[0043] AI Agent (also known as artificial intelligence agent) is an intelligent entity that can perceive the environment, make decisions and execute actions. Like humans, the intelligent agent has memory, logical analysis ability, task decomposition ability, problem decomposition ability and the ability to integrate and solve problems.
[0044] like Figure 1 As shown, an embodiment of the present invention discloses an AI digital employee training method based on deep learning, and the method comprises the following steps:
[0045] Determine a target application scenario of the AI digital employee and an associated application scenario associated with the target application scenario, and retrieve the first employee work history data corresponding to the target application scenario and the second employee work history data corresponding to the associated application scenario;
[0046] Performing feature extraction on the first employee's work history data and the second employee's work history data respectively to obtain a first question-and-answer feature set and a second question-and-answer feature set;
[0047] Using a deep learning model to align the elements in the first question-answer feature set and the second question-answer feature set, obtaining a plurality of question-answer feature pairs, each of which includes a first question-answer feature and a corresponding second question-answer feature, and forming each of the question-answer feature pairs into a training data set; wherein the first question-answer feature is associated with the first question-answer feature set, and the second question-answer feature is associated with the second question-answer feature set;
[0048] The AI digital employee is trained using the training data set so that the AI digital employee has a response capability suitable for the target application scenario.
[0049] The present invention first retrieves the first employee work history data in the target application scenario and the second employee work history data in the associated application scenario associated with the target application scenario according to the target application scenario of the AI digital employee. The application scenarios involved in the present invention are, for example, banks, hospitals, government departments, etc. Then, feature extraction is performed on the above two sets of employee work history data respectively, so as to obtain the first question and answer feature set and the second question and answer feature set. The above question and answer feature set contains the answers of real employees in the application scenario to various questions of customers. At the same time, the elements in the first question and answer feature set and the second question and answer feature set are aligned to obtain the first question and answer feature and the second question and answer feature with an associated relationship, which constitute a set of question and answer feature pairs. Finally, all the question and answer feature pairs are constructed as a training data set, and the AI digital employee is trained using the training data set so that it finally has the coping ability suitable for the target application scenario.
[0050] Therefore, on the one hand, the present invention constructs a training data set based on the correlation between the answers provided by real employees to deal with specific problems in the scene, which can better reflect that the response ability acquired by the trained AI digital employees is more matched with the corresponding scene. On the other hand, the present invention not only selects the work data of real employees in the specific target application scenarios of AI digital employees, but also further expands to the work data of real employees in related application scenarios that have a certain correlation with the target application scenarios, and combines the data of the two to extract the correlation between the answers provided by real employees to deal with specific problems in the scene, that is, the question-and-answer features, so that the trained AI digital employees have stronger response capabilities and can bring a better experience to customers.
[0051] The AI digital employee in the present invention is mainly in the form of software, but can also be embedded in various physical hardware, such as the intelligent guidance robot used in bank halls and hospital halls. The present invention does not make any limitation to this.
[0052] It should be noted that the associated application scenarios associated with the target application scenarios of the present invention may be manually specified or determined by big data analysis, and there is no specific limitation thereto.
[0053] Optionally, determining an associated application scenario associated with the target application scenario includes:
[0054] Retrieving each failure response record data corresponding to the target application scenario and performing semantic understanding on the data to obtain the scenario corresponding to each failure response record data;
[0055] Calculating the matching degree between each of the scenarios and the target application scenario, obtaining a number of target scenarios with matching degrees lower than a matching threshold, and sorting the target scenarios from large to small according to the matching degrees;
[0056] The response busyness of the target application scenario is determined, a specified number of the target scenarios ranked top are filtered out according to the response busyness, and the filtered target scenarios are determined as associated application scenarios.
[0057] In this embodiment, the associated application scenario should be an application scenario that is frequently asked by customers in the target application scenario. For example, after completing business at the bank, customers will further inquire about public transportation riding, securities, insurance, etc. Since the scenarios to which these questions belong do not match the real employees or AI digital employees (earlier versions or other brands / models) configured in the application scenario, the real employees or AI digital employees in the application scenario are often unable to respond accurately, resulting in the occurrence of failed response record data. The present invention utilizes the failed response record data, that is, retrieves each failed response record data corresponding to the target application scenario and performs semantic understanding on it, so as to obtain the scenarios to which these failed response problems correspond, removes the scenarios to which the problems with a sufficiently high degree of matching with the target application scenario belong (real employees or AI digital employees may also fail to respond to professional problems in the target application scenario), and the remaining ones are problems that do not belong to the target application scenario, and the scenarios corresponding to these remaining problems are determined as the target scenarios.
[0058] At the same time, there may be too many scenarios to which the target belongs. If all of them are added to the associated application scenarios, the amount of data of the second employee's work history data will be too large, and the subsequent training of AI digital employees will also be prone to excessive. Although AI digital employees can have stronger response capabilities, this is not economical. In this regard, the present invention evaluates the response busyness of the target application scenarios, where the unit time can be one day, half a day, etc. According to the response busyness, the target scenarios with a specified number of the highest matching degrees are screened and used as associated application scenarios. In this way, appropriate deletion of the target scenarios is achieved, which can reduce the response of AI digital employees to consulting issues with low relevance, thereby improving response efficiency.
[0059] The response busyness is mainly achieved based on the average number of consultations per unit time of customers in the target application scenario. The larger the average value, the higher the corresponding response busyness, and vice versa.
[0060] Optionally, the extracting features of the first employee's work history data and the second employee's work history data to obtain a first question-and-answer feature set and a second question-and-answer feature set respectively includes:
[0061] Extracting features from the first employee's work history data to obtain the first question-and-answer feature set;
[0062] Extracting features from the second employee's work history data to obtain a number of initial question-and-answer features, and converting each of the initial question-and-answer features into natural language to obtain an initial question-and-answer dialogue;
[0063] A BERT model is used to perform semantic understanding on each of the initial question-and-answer dialogues to obtain each of the intermediate question-and-answer dialogues, a clustering algorithm is used to cluster each of the intermediate question-and-answer features into a number of target question-and-answer dialogues, and feature extraction is performed on each of the target question-and-answer dialogues to obtain a number of target question-and-answer features, each of which constitutes the second question-and-answer feature set.
[0064] In this embodiment, since the first employee's work history data and the second employee's work history data belong to main data and auxiliary data respectively, different strategies need to be adopted when extracting features from the two.
[0065] Specifically, the feature extraction of the first employee's work history data is performed in a conventional manner, and the first question-and-answer feature set corresponding to the first employee's work history data is directly identified and retained.
[0066] The method of extracting features from the work history data of the second employee is as follows: first, extract features from the work history data of the second employee to obtain a number of initial question-and-answer features, which are of the same dimension as the question-and-answer features in the first question-and-answer feature set, namely [customer consultation question data, employee response data, label]. However, since the work history data of the second employee is auxiliary data, when the AI digital employee answers questions about related application scenarios in the target application scenario, it generally does not need to answer in a very detailed manner, but only needs to give a general answer, which can reduce disputes caused by wrong answers or lack of professionalism. In this regard, the present invention further converts the initial question-answer features obtained above into natural language to obtain an initial question-answer dialogue, and then uses the BERT model to perform semantic understanding on the initial question-answer dialogue to obtain the intermediate question-answer dialogue. The intermediate question-answer dialogue is a generalized question-answer corresponding to a specific question-answer. For example, the original question-answer is: "The actual mortgage interest rate of your bank is a bit high. What is the actual mortgage interest rate of other banks?", "The actual mortgage interest rate of XX Bank is 3.75%, and the actual mortgage interest rate of YY Bank is 4.15%." The intermediate question-answer after summary processing is: "The actual mortgage interest rate of your bank is a bit high. What is the actual mortgage interest rate of other banks?", "Generally speaking, the actual mortgage interest rates of other banks are indeed slightly lower than ours." The intermediate question-answer dialogue is then feature extracted in the above manner to obtain the target question-answer feature, and each target question-answer feature constitutes a second question-answer feature set.
[0067] Therefore, in order to reduce disputes caused by incorrect or unprofessional answers, the present invention configures AI digital employees to adopt more general answers when responding to consulting questions in non-target application scenarios. Specifically, this is achieved by performing natural language, semantic understanding and clustering, and feature re-extraction on the initial question and answer features directly extracted, which can improve the personalized response capabilities of AI digital employees that meet the actual needs of the "employer".
[0068] The above-mentioned labels are annotation data used to indicate the accuracy of the association between "customer consultation question data" and "employee response answer data", which are manually labeled or automatically labeled by machines. The labels are used to guide the subsequent training of AI digital employees.
[0069] Optionally, the using a deep learning model to align elements in the first question-answer feature set and the second question-answer feature set to obtain a plurality of groups of question-answer feature pairs includes:
[0070] The deep learning model groups the first question and answer features in the first question and answer feature set and the second question and answer features in the second question and answer feature set according to question and answer types to obtain a plurality of first groups and a plurality of second groups;
[0071] Obtaining a number of historical question-and-answer records corresponding to each of the failed response record data, calculating the average number of records of the historical question-and-answer records corresponding to all of the failed response record data, and determining the average number of records as the registration number; wherein the historical question-and-answer record is earlier in time than the failed response record data and belongs to a single question-and-answer session of the same customer;
[0072] A number of the registered number of the second groups are randomly selected to be associated with a certain first group to obtain a number of question-answer feature pairs.
[0073] In this embodiment, a deep learning model is pre-constructed and used to group the first question and answer features in the first question and answer feature set and the second question and answer features in the second question and answer feature set according to the question and answer type to obtain a number of first groups and a number of second groups. The question and answer type may be deposit type, loan type, financial management type, etc.
[0074] In addition, when customers consult the first question belonging to the target application scenario, they will also consult the second question belonging to the related application scenario, but there is a lack of a relatively stable correlation between this question and the second question and the first question. In order to ensure the response ability of AI digital employees, the existing method will choose to associate each first group with all second groups, so that the trained AI digital employees can identify the second question of the related application scenario that the customer wants to consult as quickly as possible, and be ready to answer the questions during the questioning process, which can improve the customer experience. However, this processing method will seriously increase the difficulty of training, and will make the response topology network of AI digital employees very complex, which may reduce the response efficiency.
[0075] In response to the above problems, the present invention obtains several historical question and answer records corresponding to the aforementioned response failure record data. The historical question and answer records are questions asked by specific customers in a single consultation process that belong to the target application scenario, and then calculates the average number of historical question and answer records corresponding to each response failure record data. The average number of records represents the general correspondence between the response failure situation and the number of questions in the target application scenario. Then, a second grouping of the average number of records is randomly selected to associate with the aforementioned first grouping, that is, to form a new grouping, which contains a group of question and answer feature pairs (including both question and answer pairs of the target application scenario and question and answer pairs of the associated application scenario). By iteratively processing in the above manner, a number of groups of question and answer feature pairs corresponding to the number of the first grouping are obtained.
[0076] Of course, the second grouping may also be used as a benchmark, that is, a number of first groups with a certain number of registrations may be randomly selected to be associated with a second grouping to obtain a set of question-answer feature pairs.
[0077] Optionally, the deep learning model groups the first question and answer features in the first question and answer feature set and the second question and answer features in the second question and answer feature set according to question and answer types, including:
[0078] The deep learning model retrieves an operation specification file corresponding to the target application scenario, and determines and obtains a plurality of question and answer types according to the operation specification file;
[0079] The first question and answer features in the first question and answer feature set and the second question and answer features in the second question and answer feature set are grouped according to the question and answer type.
[0080] In this embodiment, the trained deep learning model can classify the first question-and-answer feature and the second question-and-answer feature separately by their respective characteristics, and then realize grouping. However, when the distinction between the business types in the target application scenario is not obvious enough, the deep learning model will easily group multiple question-and-answer features with similar business types into one group, which does not meet the actual needs of the target application scenario. Therefore, the present invention further obtains the operation specification file corresponding to the target application scenario, and uses the operation specification file to guide the classification and grouping of the deep learning model, so as to obtain a grouping suitable for the target application scenario.
[0081] Among them, the deep learning model in the present invention is preferably built and trained based on Transformer.
[0082] Optionally, the determining of a plurality of question and answer types according to the operation specification file includes:
[0083] The deep learning model parses the directory part in the operation specification file, determines the configuration quantity of the secondary blocks corresponding to each primary block, and calculates the average configuration quantity of each configuration quantity;
[0084] If the average configuration quantity exceeds a threshold, the deep learning model is set to determine and derive a number of the question and answer types according to the operation specification file; otherwise, the deep learning model is set not to determine and derive a number of the question and answer types according to the operation specification file.
[0085] In this embodiment, the directory part in the operation specification file will record the classification information for various business types, but the degree of detail of the classification is different in different application scenarios. The present invention determines whether to put the operation specification file into the deep learning model for grouping guidance based on the average configuration number of the secondary sections corresponding to the main sections of the directory part. Specifically, when the average configuration number exceeds the threshold, it means that the business type of the target application scenario is divided very finely. At this time, the operation specification file is needed to guide the grouping operation of the deep learning model. Otherwise, it means that the business type of the target application scenario is divided very roughly, and the deep learning model can easily distinguish various business types. At this time, the guidance of the operation specification file is not needed.
[0086] It should be noted that the main sections of the directory part of the operating specification document are, for example, "1.1 Loan issues", "1.2 Deposit issues", "1.3 Financial management issues", and the secondary sections are, for example, "1.1.1 Lending issues", "1.2.2 Fixed deposit issues", etc. The configuration quantity refers to the number of setting levels of the secondary sections. For example, "1.2.2" represents three levels, that is, the configuration quantity is 3, and "1.2.2.1" represents four levels, that is, the configuration quantity is 4.
[0087] like Figure 2 As shown, the present invention also discloses an AI digital employee training system based on deep learning, the system comprising a retrieval module, a feature extraction module, a training set building module, and a training module;
[0088] The retrieval module is used to determine the target application scenario of the AI digital employee and the associated application scenario associated with the target application scenario, and retrieve the first employee work history data corresponding to the target application scenario and the second employee work history data corresponding to the associated application scenario;
[0089] The feature extraction module is used to extract features from the first employee's work history data and the second employee's work history data respectively to obtain a first question-answer feature set and a second question-answer feature set;
[0090] The training set building module is used to use a deep learning model to align the elements in the first question-answer feature set and the second question-answer feature set to obtain a plurality of question-answer feature pairs, each of which contains a first question-answer feature and a corresponding second question-answer feature, and to build each of the question-answer feature pairs into a training data set; wherein the first question-answer feature is associated with the first question-answer feature set, and the second question-answer feature is associated with the second question-answer feature set;
[0091] The training module is used to train the AI digital employee using the training data set so that the AI digital employee has a response capability suitable for the target application scenario.
[0092] The present invention also discloses an electronic device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program implements the method described in any of the preceding items when executed by the processor.
[0093] The present invention also discloses a computer storage medium, which stores a computer program that can be executed by a processor to implement any of the methods described in the preceding items.
[0094] The present invention also discloses a computer program product, which includes a computer program that can be executed by a processor to implement any of the methods described in the preceding items.
[0095] It should be understood that the terms used in the present application are only for the purpose of describing specific embodiments, but are not intended to limit the present application. The singular forms of "a", "said" and "the" used in the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in this article refers to and includes any or all possible combinations of one or more associated listed items. The expressions "having", "may have", "include" and "include", or "may include" and "may include" in this application can be used to indicate the presence of corresponding features (e.g., elements such as numerical values, functions, operations or components), but the presence of additional features is not excluded.
[0096] It should be understood that although the terms first, second, third, etc. may be used in this application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other, and are not necessarily used to describe a specific order or sequence. For example, without departing from the scope of the present invention, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. The technical solutions recorded in the embodiments of the present application can be arbitrarily combined without conflict. In several embodiments provided in this application, it should be understood that the disclosed methods, devices and clothing processing equipment can be implemented in other ways. The embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the clothing processing equipment or units can be electrical, mechanical or other forms.
[0097] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.
[0098] In addition, all functional units in the embodiments of the present application can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the above integrated units can be implemented in the form of hardware or in the form of hardware plus software functional units. The above is only a specific implementation method of the present application, but the protection scope of the present application is not limited to this. Any technician familiar with the technical field can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application.
Claims
1. A deep learning-based AI digital employee training method, characterized by: The method comprises the following steps: Determine a target application scenario of the AI digital employee and an associated application scenario associated with the target application scenario, and retrieve the first employee work history data corresponding to the target application scenario and the second employee work history data corresponding to the associated application scenario; Performing feature extraction on the first employee's work history data and the second employee's work history data respectively to obtain a first question-and-answer feature set and a second question-and-answer feature set; Using a deep learning model to align the elements in the first question-answer feature set and the second question-answer feature set, obtaining a plurality of question-answer feature pairs, each of which includes a first question-answer feature and a corresponding second question-answer feature, and forming each of the question-answer feature pairs into a training data set; wherein the first question-answer feature is associated with the first question-answer feature set, and the second question-answer feature is associated with the second question-answer feature set; Using the training data set to train the AI digital employee so that the AI digital employee has a coping capability suitable for the target application scenario; Determining an associated application scenario associated with the target application scenario includes: Retrieving each failure response record data corresponding to the target application scenario and performing semantic understanding on the data to obtain the scenario corresponding to each failure response record data; Calculating the matching degree between each of the scenarios and the target application scenario, obtaining a number of target scenarios with matching degrees lower than a matching threshold, and sorting the target scenarios from large to small according to the matching degrees; Determine the response busyness of the target application scenario, filter a specified number of the target scenarios that are ranked top according to the response busyness, and determine the filtered target scenarios as associated application scenarios; wherein the response busyness is implemented based on the average value of the number of consultations per unit time by customers in the target application scenario, and the larger the average value, the higher the corresponding response busyness, and vice versa.
2. The AI digital employee training method based on deep learning according to claim 1, characterized in that: The extracting features of the first employee's work history data and the second employee's work history data to obtain a first question-and-answer feature set and a second question-and-answer feature set respectively includes: Extracting features from the first employee's work history data to obtain the first question-and-answer feature set; Extracting features from the second employee's work history data to obtain a number of initial question-and-answer features, and converting each of the initial question-and-answer features into natural language to obtain an initial question-and-answer dialogue; A BERT model is used to perform semantic understanding on each of the initial question-and-answer dialogues to obtain each of the intermediate question-and-answer dialogues, a clustering algorithm is used to cluster each of the intermediate question-and-answer features into a number of target question-and-answer dialogues, and feature extraction is performed on each of the target question-and-answer dialogues to obtain a number of target question-and-answer features, each of which constitutes the second question-and-answer feature set.
3. The AI digital employee training method based on deep learning according to claim 2 is characterized in that: The using a deep learning model to align the elements in the first question-answer feature set and the second question-answer feature set to obtain a plurality of groups of question-answer feature pairs includes: The deep learning model groups the first question and answer features in the first question and answer feature set and the second question and answer features in the second question and answer feature set according to question and answer types to obtain a plurality of first groups and a plurality of second groups; Obtaining a number of historical question-and-answer records corresponding to each of the failed response record data, calculating the average number of records of the historical question-and-answer records corresponding to all of the failed response record data, and determining the average number of records as the registration number; wherein the historical question-and-answer record is earlier in time than the failed response record data and belongs to a single question-and-answer session of the same customer; A number of the registered number of the second groups are randomly selected to be associated with a certain first group to obtain a number of question-answer feature pairs.
4. The AI digital employee training method based on deep learning according to claim 3 is characterized by: The deep learning model groups the first question and answer features in the first question and answer feature set and the second question and answer features in the second question and answer feature set according to question and answer types, including: The deep learning model retrieves an operation specification file corresponding to the target application scenario, and determines and obtains a plurality of question and answer types according to the operation specification file; The first question and answer features in the first question and answer feature set and the second question and answer features in the second question and answer feature set are grouped according to the question and answer type.
5. The AI digital employee training method based on deep learning according to claim 4 is characterized in that: The several question and answer types are determined according to the operation specification document, including: The deep learning model parses the directory part in the operation specification file, determines the configuration quantity of the secondary blocks corresponding to each primary block, and calculates the average configuration quantity of each configuration quantity; If the average configuration quantity exceeds a threshold, setting the deep learning model to determine and obtain a number of the question and answer types according to the operation specification file; Otherwise, the deep learning model is set not to determine and derive a number of the question and answer types according to the operation specification file.
6. An AI digital employee training system based on deep learning, the system is based on the method according to any one of claims 1 to 5, the system comprises a retrieval module, a feature extraction module, a training set building module, and a training module; characterized in that: The retrieval module is used to determine the target application scenario of the AI digital employee and the associated application scenario associated with the target application scenario, and retrieve the first employee work history data corresponding to the target application scenario and the second employee work history data corresponding to the associated application scenario; The feature extraction module is used to extract features from the first employee's work history data and the second employee's work history data respectively to obtain a first question-answer feature set and a second question-answer feature set; The training set building module is used to use a deep learning model to align the elements in the first question-answer feature set and the second question-answer feature set to obtain a plurality of question-answer feature pairs, each of which contains a first question-answer feature and a corresponding second question-answer feature, and to build each of the question-answer feature pairs into a training data set; wherein the first question-answer feature is associated with the first question-answer feature set, and the second question-answer feature is associated with the second question-answer feature set; The training module is used to train the AI digital employee using the training data set so that the AI digital employee has a response capability suitable for the target application scenario.
7. An electronic device, characterized in that: The electronic device comprises: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program implements the method according to any one of claims 1 to 5 when executed by the processor.
8. A computer storage medium, characterized in that: The computer storage medium stores a computer program that can be executed by a processor to implement the method according to any one of claims 1 to 5.
9. A computer program product, characterized in that: The computer program product comprises a computer program executable by a processor to implement the method according to any one of claims 1 to 5.
