Training method of dialogue model, artificial intelligence interview method, computing device, storage medium and program product

Abstract

Embodiments of the present application provide a conversation model training method, an artificial intelligence interview method, a computing device, a computer readable storage medium and a computer program product. The conversation model training method comprises: determining a first conversation model and a second conversation model; taking the first conversation model as a student model and taking the second conversation model as a teacher model; obtaining first sample data; inputting the first sample data into the student model; splicing target element data and the first sample data to obtain second sample data; inputting the second sample data into the student model and the teacher model respectively; determining difference information of first probability distribution and second probability distribution corresponding to different element data respectively; and adjusting the student model based on the difference information to obtain a target conversation model. The technical solution provided by the embodiments of the present application reduces the training difficulty of the student model and improves the performance of the target conversation model generated by training.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method for training a dialogue model, an artificial intelligence interview method, a computing device, a computer-readable storage medium, and a computer program product. Background Technology

[0002] In today's digital age, artificial intelligence technology is being used more and more widely in the recruitment field, especially AI (Artificial Intelligence) interview functions, which have brought great convenience to both companies and job seekers.

[0003] AI interviews typically utilize dialogue models. These models can ask questions to job seekers and, after receiving their responses, continue to ask questions based on those responses, thus completing the interview through dialogue.

[0004] In related technologies, knowledge distillation can be used to train dialogue models. Simply put, knowledge distillation is a model compression and transfer learning technique. Its core idea is to transfer the knowledge of a complex, high-performance teacher model to a relatively simple student model, i.e., a dialogue model.

[0005] In realizing the concept of this application, the inventors discovered that in related technologies, the training method based on knowledge distillation processes sample data separately by a teacher model and a student model to obtain their respective probability distributions for different element data, and then trains the student model based on the differences in probability distributions. This requires the student model to strictly fit the output distribution of the teacher model, but this forced fitting may exceed the learning capacity of the student model, resulting in poor performance of the trained dialogue model. Summary of the Invention

[0006] This application provides a method for training a dialogue model, an artificial intelligence interview method, a computing device, a computer-readable storage medium, and a computer program product.

[0007] Firstly, embodiments of this application provide a method for training a dialogue model, including:

[0008] A first dialogue model and a second dialogue model are determined, wherein the number of parameters in the second dialogue model is greater than the number of parameters in the first dialogue model.

[0009] Use the first dialogue model as the student model and the second dialogue model as the teacher model;

[0010] Obtain first sample data; the first sample data includes at least one round of sample dialogue data;

[0011] The first sample data is input into the student model to perform inference operations based on the first sample data, calculate the first probability distribution of different element data, and select target element data according to the first probability distribution; the element data is the smallest data unit.

[0012] The target element data is concatenated with the first sample data to obtain the second sample data;

[0013] The second sample data is input into the student model and the teacher model respectively, so that the student model performs inference operations based on the second sample data to calculate the first probability distribution of different element data, and the teacher model performs inference operations based on the second sample data to calculate the second probability distribution of different element data.

[0014] Determine the differences between the first probability distribution and the second probability distribution corresponding to different element data;

[0015] The student model is adjusted based on the difference information to transfer the knowledge of the second dialogue model to the first dialogue model, thereby obtaining a target dialogue model; the target dialogue model is used to generate target question data from a list of questions corresponding to the target position by combining at least one round of historical dialogue data.

[0016] Secondly, this application provides an artificial intelligence interview method applied to a server, the method comprising:

[0017] Obtain the interview request sent by the client; the interview request is generated by the client in response to the interview operation triggered by the target user for the target position;

[0018] Determine the question list corresponding to the target position; the question list includes multiple question data.

[0019] A target dialogue model is determined; the target dialogue model is obtained by transferring knowledge from a second dialogue model to a first dialogue model; the first dialogue model is used as a student model, and the second dialogue model is used as a teacher model; the student model is adjusted based on the difference information between a first probability distribution and a second probability distribution of different element data generated through reasoning; the first probability distribution is generated by the student model based on second sample data; the second probability distribution is generated by the teacher model based on second sample data; the second sample input model is obtained by concatenating the first sample data with the target element data; the target element data is generated by the student model based on the first sample data.

[0020] Using the target dialogue model combined with at least one round of historical dialogue data, target question data is generated from the question list;

[0021] The target problem data is sent to the client.

[0022] Obtain the response data from the client and return to the step of generating target question data from the question list by combining the target dialogue model with at least one round of historical dialogue data to continue execution.

[0023] Thirdly, this application provides a training apparatus for a dialogue model, comprising:

[0024] The first determining module is used to determine the first dialogue model and the second dialogue model, wherein the number of parameters of the second dialogue model is greater than the number of parameters of the first dialogue model.

[0025] The second determining module is used to use the first dialogue model as a student model and the second dialogue model as a teacher model.

[0026] A sample acquisition module is used to acquire first sample data; the first sample data includes at least one round of sample dialogue data.

[0027] The first input module is used to input the first sample data into the student model, so as to use the student model to perform inference operations based on the first sample data, to calculate the first probability distribution of different element data, and to select target element data according to the first probability distribution; the element data is the smallest data unit.

[0028] The splicing module is used to splice the target element data with the first sample data to obtain the second sample data;

[0029] The second input module is used to input the second sample data into the student model and the teacher model respectively, so as to use the student model to perform inference operations based on the second sample data to calculate the first probability distribution of different element data, and use the teacher model to perform inference operations based on the second sample data to calculate the second probability distribution of different element data.

[0030] The difference determination module is used to determine the difference information between the first probability distribution and the second probability distribution corresponding to different element data.

[0031] The training module is used to adjust the student model based on the difference information to transfer the knowledge of the second dialogue model to the first dialogue model and obtain the target dialogue model; the target dialogue model is used to generate target question data from the question list corresponding to the target position by combining at least one round of historical dialogue data.

[0032] Fourthly, this application provides an artificial intelligence interviewing device applied to a server, the method comprising:

[0033] The first request acquisition module is used to acquire interview requests sent by the client; the interview request is generated by the client in response to the interview operation triggered by the target user for the target position;

[0034] The first question list determination module is used to determine the question list corresponding to the target position; the question list includes multiple question data.

[0035] A dialogue model determination module is used to determine a target dialogue model. The target dialogue model is obtained by transferring knowledge from a second dialogue model to a first dialogue model. The first dialogue model serves as a student model, and the second dialogue model serves as a teacher model. The student model is adjusted based on the difference information between a first probability distribution and a second probability distribution of different element data generated through reasoning. The first probability distribution is generated by the student model based on second sample data. The second probability distribution is generated by the teacher model based on second sample data. The second sample input model is obtained by concatenating first sample data with target element data. The target element data is generated by the student model based on the first sample data.

[0036] The first question generation module is used to generate target question data from the question list by combining the target dialogue model with at least one round of historical dialogue data.

[0037] The first question sending module is used to send the target question data to the client;

[0038] The first response acquisition module is used to acquire the response data fed back by the client and return to the step of generating target question data from the question list by combining the target dialogue model with at least one round of historical dialogue data to continue execution.

[0039] Fifthly, this application provides a computing device, including a processing component and a storage component;

[0040] The storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement the dialogue model training method provided in the embodiments of this application, or to implement the artificial intelligence interview method provided in the embodiments of this application.

[0041] Sixthly, this application provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processing component, it implements the training method for the dialogue model provided in this application embodiment, or implements the artificial intelligence interview method provided in this application embodiment.

[0042] In a seventh aspect, this application provides a computer program product, including a computer program / instruction, which, when executed by a processing component, implements the training method for the dialogue model provided in this application embodiment, or implements the artificial intelligence interview method provided in this application embodiment.

[0043] This embodiment first determines a first dialogue model and a second dialogue model with different parameter values. The first dialogue model is used as the student model, and the second dialogue model as the teacher model. Next, first sample data containing at least one round of sample dialogue data is obtained and input into the student model for inference operations. A first probability distribution of different element data is calculated, and target element data is selected. The target element data is concatenated with the first sample data to obtain second sample data. Then, the second sample data is input into both the student and teacher models respectively, and the first and second probability distributions of different element data are calculated for each. The difference information between the first and second probability distributions of different element data is then determined. Finally, based on the difference information, the student model is adjusted, transferring knowledge from the second dialogue model to the first dialogue model to obtain the target dialogue model. By comparing the differences between the outputs of the student model and the teacher model after the student model has already produced its own output, and guiding the student model to adjust in a better direction based on the difference information, the training difficulty of the student model is reduced, and the performance of the generated target dialogue model is improved.

[0044] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 A system architecture diagram illustrating a technical solution of an embodiment of this application that can be applied thereto is shown;

[0047] Figure 2 A flowchart illustrating a training method for a dialogue model according to an embodiment of this application is shown;

[0048] Figure 3 A flowchart of an artificial intelligence interview method provided in one embodiment of this application is shown;

[0049] Figure 4 A schematic diagram of the artificial intelligence interview method provided in an embodiment of this application is shown;

[0050] Figure 5 A block diagram of a training apparatus for a dialogue model according to an embodiment of this application is shown;

[0051] Figure 6 A block diagram of an artificial intelligence interview device provided in an embodiment of this application is shown;

[0052] Figure 7 A block diagram of a computing device provided in one embodiment of this application is shown. Detailed Implementation

[0053] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0054] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0055] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0056] It should be noted that the technical solutions of this application embodiment are applicable to the network virtual environment. The user described generally refers to a "virtual user". Real users can register user accounts on the server through registration to obtain user identity in the network environment. In this application embodiment, the same user account can be used to log in to the server through different types of clients, so that the server can identify the same user. Of course, different user accounts can also be used to log in to the server through different types of clients. The server stores different user account binding relationships, so different user accounts with binding relationships can be considered as the same user.

[0057] In today's digital age, artificial intelligence technology is being used more and more widely in the recruitment field, especially AI (Artificial Intelligence) interview functions, which have brought great convenience to both companies and job seekers.

[0058] AI interviews typically utilize dialogue models. These models can ask questions to job seekers and, after receiving their responses, continue to ask questions based on those responses, thus completing the interview through dialogue.

[0059] In related technologies, knowledge distillation based on positive KL divergence can be used to train dialogue models. Knowledge distillation, simply put, is a model compression and transfer learning technique. Its core idea is to transfer the knowledge of a complex, high-performance teacher model to a relatively simple student model, i.e., the dialogue model.

[0060] In training using positive KL divergence, sample data can be processed separately by the teacher model and the student model to obtain their respective probability distributions for different data elements. We can denote the probability distribution of the teacher model as P(x) and that of the student model as Q(x). Then, for each data element, the positive KL divergence can be calculated based on P(x) and Q(x). The positive KL divergence value quantifies the difference between P(x) and Q(x); a larger value indicates a greater difference between the two distributions. During model training, the positive KL divergence value can be used as a loss function. By minimizing the positive KL divergence value, the parameters of the student model can be adjusted to make the output distribution of the student model as close as possible to the output distribution of the teacher model.

[0061] In realizing the concept of this application, the inventors discovered that when there is a significant difference in the data output distribution between the student model and the teacher model, the positive KL divergence makes learning extremely difficult for the student model. This is because it requires the student model to strictly fit the output distribution of the teacher model, and such forced fitting may exceed the learning capacity of the student model. For example, in complex natural language processing tasks, the teacher model may have trained a very detailed output distribution based on large-scale data, but the student model, due to its own structure and relatively limited training data, finds it difficult to directly learn such a complex distribution, thus hindering the learning process.

[0062] To address the technical problems existing in related technologies, the embodiments of this application first determine a first dialogue model and a second dialogue model with different parameter values. The first dialogue model is used as the student model, and the second dialogue model is used as the teacher model. Next, first sample data containing at least one round of sample dialogue data is obtained and input into the student model for inference operations. A first probability distribution of different element data is calculated, and target element data is selected. The target element data is concatenated with the first sample data to obtain second sample data. Then, the second sample data is input into both the student model and the teacher model, and the first and second probability distributions of different element data are calculated respectively. The difference information between the first and second probability distributions of different element data is then determined. Finally, the student model is adjusted based on the difference information, transferring the knowledge of the second dialogue model to the first dialogue model to obtain the target dialogue model. Through the technical solution provided by this application embodiment, training can be performed based on the output of the student model. Unlike in positive KL divergence, the teacher model does not provide a probability distribution for the student model to fit. Instead, after the student model has its own output, the differences between the two are compared, guiding the student model to adjust in a better direction. This reduces the training difficulty of the student model and improves the performance of the generated target dialogue model.

[0063] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0064] Figure 1 A system architecture diagram is shown that can be applied to a technical solution of an embodiment of this application. The system architecture may include a client 101 and a server 102.

[0065] In this configuration, client 101 and server 102 establish a connection via a network. The network provides a communication link between client 101 and server 102. The network can include various connection types, such as wired or wireless communication links or fiber optic cables. Optionally, the server can communicate with the client via a mobile network. Optionally, the client can also establish a communication connection with the server via Bluetooth, WiFi, infrared, etc.

[0066] Client 101 can interact with server 102 via the network to receive or send messages, etc.

[0067] The client 101 can be a browser, an app (application), a web application such as an H5 (HyperText Markup Language 5) application, a lightweight application (also known as a mini-program), or a cloud application. The client 101 can be deployed on electronic devices and depends on the device to run or on certain apps within the device. Electronic devices can have displays and support information browsing, such as personal mobile terminals like mobile phones, tablets, personal computers, desktop computers, smart speakers, smartwatches, etc. For ease of understanding, Figure 1 The client is primarily represented by the image of a device. Various other types of applications can also be configured in electronic devices, such as human-computer interaction applications, model training applications, text processing applications, web browser applications, shopping applications, search applications, instant messaging tools, email clients, and social media platform software. An electronic device can refer to a user-used device with the computing, internet access, and communication functions required by the user, such as a mobile phone, tablet computer, personal computer, or wearable device. An electronic device typically includes at least one processing component and at least one storage component. Electronic devices may also include basic configurations such as network interface cards (NICs), I / O buses, and audio / video components; this application does not limit this. Optionally, depending on the implementation of the electronic device, it may also include some peripheral devices, such as a keyboard, mouse, input pen, and printer; this application does not limit this.

[0068] Server 102 may include servers that provide various services, such as a server for background training that supports the model used on client 101, or a server that processes interactive information sent by the client.

[0069] It should be noted that server 102 can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. The server can also be a server in a distributed system, or a server integrated with blockchain. The server can also be a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0070] It should be noted that the dialogue model training method and AI interview method provided in the embodiments of this application are generally executed by the server 102, and the corresponding dialogue model training device and AI interview device are generally set in the server 102. However, in other embodiments of this application, the client 101 may also have similar functions to the server 102, thereby executing the dialogue model training method and AI interview method provided in the embodiments of this application. In other embodiments, the dialogue model training method and AI interview method provided in the embodiments of this application may also be jointly executed by the client 101 and the server 102.

[0071] It should be understood that Figure 1 The number of clients and servers shown is merely illustrative. Depending on implementation needs, there can be any number of clients and servers.

[0072] The implementation details of the technical solutions in the embodiments of this application are described in detail below.

[0073] Figure 2 The flowchart illustrates a training method for a dialogue model according to an embodiment of this application, as shown below. Figure 2 As shown, the method may specifically include the following steps:

[0074] 201: Determine the first dialogue model and the second dialogue model, wherein the second dialogue model has more parameters than the first dialogue model.

[0075] 202: Use the first dialogue model as the student model and the second dialogue model as the teacher model.

[0076] The first dialogue model can be a model with relatively few parameters that is more focused on a specific dialogue task. In the embodiments of this application, the specific dialogue task can include a dialogue task in an interview scenario.

[0077] Secondary dialogue models can have a larger number of parameters, thus they can be more complex and powerful models that can be trained on a wider range of datasets, possessing richer knowledge and stronger processing capabilities.

[0078] In embodiments of this application, knowledge distillation can be used to train the first dialogue model. Within the framework of knowledge distillation, the first dialogue model can be viewed as a student model, which can learn knowledge and skills from another, more powerful model, namely the teacher model. In embodiments of this application, a second dialogue model with a larger number of parameters can be used as the teacher model.

[0079] Through knowledge distillation, the teacher model can leverage its rich parameters and extensive training knowledge to guide the student model, thereby improving the student model's performance on specific tasks while maintaining its own advantages of relatively simple structure, such as faster inference speed and lower resource consumption.

[0080] 203: Obtain the first sample data; the first sample data includes at least one round of sample dialogue data.

[0081] The first sample data may include at least one round of sample dialogue data, and the at least one round of sample dialogue data may be arranged in the order of the dialogue.

[0082] In one possible implementation, at least one round of sample dialogue data can be obtained by collecting dialogue data generated during at least one historical interview process. This sample dialogue data can record the conversation between the interviewer and the job seeker in a real interview scenario.

[0083] In real-world dialogue scenarios, the order of conversations affects semantic understanding and logical coherence. For example, in an interview, the interviewer might ask follow-up questions based on the job seeker's previous answers, or adjust the direction and depth of the questions as the interview progresses. This sequential nature allows the first dialogue model to learn the natural flow of the conversation and the relevance of the context.

[0084] 204: Input the first sample data into the student model to perform inference operations based on the first sample data, calculate the first probability distribution of different element data, and select the target element data according to the first probability distribution; the element data is the smallest data unit.

[0085] After obtaining the first sample data, it can be input into the student model.

[0086] After receiving the first sample data, the student model can perform inference operations. In embodiments of this application, the inference operation can refer to the process by which the student model processes the first sample data based on its internal neural network structure and training parameters to calculate the first probability distribution of different element data.

[0087] In embodiments of this application, element data can be the smallest data unit; for example, in a natural language processing task, element data can be a single character, word, morpheme, or other smallest semantic unit. The student model can calculate a first probability distribution of this element data.

[0088] The first probability distribution represents the likelihood of each element appearing in the current input. For example, in a dialogue generation task, if the element is a word, the student model can calculate the probability distribution of the next possible word based on the first sample data. For instance, assuming the student model's input is "Please describe your role in the project," the model might calculate a series of probability distributions for words, such as "experience" with a probability of 0.3, "role" with a probability of 0.25, "contribution" with a probability of 0.2, and so on. This probability distribution is based on the language patterns and semantic relationships learned by the student model during training, reflecting what the student model believes the next element might be given the current input.

[0089] Based on the calculated first probability distribution, the student model can select target element data. When selecting target element data from different data types, the student model can employ strategies such as a greedy strategy or a random sampling strategy. In a greedy strategy, the student model can choose the element data with the highest probability as the target element data. For example, in the example above, if a greedy strategy is used, the student model can choose "experience" as the target element data because it has the highest probability of 0.3. In a random sampling strategy, the student model can randomly sample according to the probability distribution, thus increasing the diversity of the generated content.

[0090] 205: Concatenate the target element data with the first sample data to obtain the second sample data.

[0091] By concatenating the target element data with the first sample data, a continuous dialogue process can be simulated. In natural language processing tasks, especially in scenarios involving dialogue generation, the model needs to continuously update its input based on previous content in order to continue generating subsequent content.

[0092] For example, concatenating the target element data with the first sample data to obtain the second sample data can be done through the following process:

[0093] Suppose the first sample data is a dialogue sequence, for example, in an interview scenario. The first sample data might be: "Interviewer: Please describe your experience in the project. Applicant: I participated in a software project, ". Based on the previous inference operations, the student model can select the target element data from the calculated first probability distribution, which might be "development". By concatenating the target element data to the end of the first sample data, the second sample data is obtained. At this point, the second sample data becomes: "Interviewer: Please describe your experience in the project. Applicant: I participated in a software project, development."

[0094] By adding newly generated element data to the original sample data to form new input (i.e., second sample data), the model can be provided with updated information, enabling it to continue learning and reasoning based on the updated input.

[0095] After generating the second sample data, it can be re-input into the student and teacher models, allowing them to further calculate probability distributions and perform new inference operations based on the updated information. This helps the models learn the coherence and logic of dialogues or texts, enabling them to gradually improve their performance through continuous iteration, better grasp the structure and semantics of language, and lay the foundation for transferring knowledge from the teacher model to the student model and generating high-quality dialogues or texts.

[0096] 206: Input the second sample data into the student model and the teacher model respectively, so that the student model can perform inference operations based on the second sample data to calculate the first probability distribution of different element data, and the teacher model can perform inference operations based on the second sample data to calculate the second probability distribution of different element data.

[0097] By inputting the second sample data into the student model and the teacher model respectively, the student model and the teacher model can process and analyze the input second sample data according to their respective parameters and structures, thereby calculating the probability distribution of different element data.

[0098] 207: Determine the differences between the first probability distribution and the second probability distribution corresponding to different element data.

[0099] After inputting the second sample data, the student model and the teacher model may each calculate the probability distribution of the next word's appearance. By comparing these two probability distributions, differences between the student model's predictions and the teacher model's predictions in certain situations can be identified. This allows for adjustments to the student model's parameters, enabling it to predict the next word more accurately. In this way, the student model can learn more accurate language patterns and knowledge from the teacher model, improving the quality and accuracy of its generated text.

[0100] 208: Adjust the student model based on the difference information to transfer the knowledge of the second dialogue model to the first dialogue model to obtain the target dialogue model; the target dialogue model is used to generate target question data from the question list corresponding to the target position by combining at least one round of historical dialogue data.

[0101] After identifying the differences, the training parameters of the student model can be adjusted so that the first probability distribution output by the student model is closer to the second probability distribution output by the teacher model. In this way, the student model can continuously learn from the performance of the teacher model when processing the same input, learn better knowledge and experience from the teacher model, and thus transfer the knowledge of the teacher model to the student model.

[0102] After multiple adjustments and optimizations, the final student model can be considered the target dialogue model. This target dialogue model combines the structure of the student model with some knowledge from the teacher model. When handling tasks such as interview dialogues, it can leverage the knowledge advantages of the teacher model while maintaining the characteristics of the student model. For example, the target dialogue model may inherit the teacher model's abilities in language patterns, semantic understanding, and logical reasoning, while also possessing advantages in resource utilization and computational efficiency, similar to the student model. It is suitable for application in practical dialogue systems, such as intelligent interview systems, which can generate more appropriate and fluent target question data from a list of questions corresponding to the target position based on at least one round of historical dialogue data.

[0103] In some embodiments, the method may further include:

[0104] Acquire pre-training data, which includes third sample data and sample question data corresponding to the third sample data; the third sample data includes at least one round of sample dialogue data; each round of sample dialogue data includes sample question data and sample response data corresponding to the question data.

[0105] The first dialogue model and the second dialogue model were pre-trained using pre-trained data.

[0106] By inputting the acquired pre-trained data into the first and second dialogue models, these models can learn language patterns, semantic understanding, and logical relationships within the dialogue based on sample question data and corresponding sample response data from the pre-trained data. For example, they can learn common answer patterns for different questions and how to respond appropriately based on context. Through continuous processing and learning from the pre-trained data, the first and second dialogue models can gradually adjust their parameters to adapt to the requirements of this dialogue task, thereby improving their capabilities in dialogue-related tasks.

[0107] In the application embodiment, a first dialogue model and a second dialogue model with different parameter values ​​are first determined. The first dialogue model is used as the student model, and the second dialogue model is used as the teacher model. Next, first sample data containing at least one round of sample dialogue data is obtained and input into the student model for inference operations. The first probability distribution of different element data is calculated, and target element data is selected. The target element data is concatenated with the first sample data to obtain second sample data. Then, the second sample data is input into both the student model and the teacher model, and the first and second probability distributions of different element data are calculated respectively. The difference information between the first and second probability distributions of different element data is then determined. Finally, the student model is adjusted based on the difference information, transferring knowledge from the second dialogue model to the first dialogue model to obtain the target dialogue model. By comparing the difference information between the outputs of the student model and the teacher model after the student model has already produced its own output, and guiding the student model to adjust in a better direction based on the difference information, the training difficulty of the student model is reduced, and the performance of the generated target dialogue model is improved.

[0108] In some embodiments, pre-training the first dialogue model and the second dialogue model using pre-trained data can be specifically implemented as follows:

[0109] For either the first dialogue model or the second dialogue model, the third sample data is used as the input data of the model, and the sample question data is used as the training label.

[0110] The model is trained using the input data and training labels.

[0111] When either the first or second dialogue model receives the third sample data as input, this third sample data can be propagated forward within the model's neural network structure. Thus, the model can perform feature extraction, semantic understanding, and other operations on the input data based on its own parameters and structure, ultimately generating an output. This output may include a prediction of the next possible dialogue content or a preliminary answer to a related question.

[0112] After the model generates output, the output can be compared with the training labels to calculate the difference between the model's output and the training labels using a loss function. Loss functions can include, for example, the cross-entropy loss function, which measures the distance between the probability distribution of the model's output and the true labels. A higher loss value indicates a larger difference between the model's output and the training labels, and vice versa.

[0113] Based on the loss value calculated using the loss function, the gradient of the loss function with respect to the model parameters can be calculated using the backpropagation algorithm. The gradient represents the direction and extent of the influence of parameter changes on the loss value. Then, optimization algorithms (such as stochastic gradient descent, Adam, etc.) are used to update the model parameters based on the gradient, enabling the model to produce an output closer to the training labels when encountering similar inputs in the future. This process is repeated continuously, with multiple iterations of training to gradually adjust the model parameters, thereby continuously improving the model's performance and ultimately enabling it to better handle dialogue-related tasks.

[0114] In some embodiments, obtaining the first sample data can be specifically implemented as follows:

[0115] Determine at least one round of sample dialogue data, and the required question output;

[0116] The first sample data is generated based on at least one round of sample dialogue data and the question output requirements.

[0117] The output requirements can be specifications and expectations for the model's output, clarifying the conditions that the model's output must meet after processing the output from the sample dialogue data. In interview question generation scenarios, output requirements might include, for example, the language style of the question (formal, concise, etc.), the type of question (technical, behavioral, etc.), the relevance of the question to the interview position, and the difficulty level of the question. Furthermore, output requirements can also include requirements for the model's capabilities; for example, the model could be required to select one question from a pre-defined list in sequence for output.

[0118] The first sample data can be implemented as a prompt word. A prompt word is an input format used to prompt or guide the model to give the expected output. A prompt word is a natural language input, similar to a command or instruction, to let the model know what it needs to do.

[0119] In this embodiment of the application, in order to facilitate the generation of prompt information, a prompt template can be pre-configured. By adding at least one round of sample dialogue data and the question output requirements to the prompt template, the corresponding prompt words can be generated.

[0120] The prompt template can include the following parts:

[0121] Sample dialogue data from at least one round: Interviewer: "Please share the responsibilities you had in a recent project."

[0122] Job seeker: "I mainly need XXXX."

[0123] Output requirements: Generate a formal, professional question that addresses the job seeker's technical skills.

[0124] In some embodiments, the second sample data is input into both the student model and the teacher model, respectively, so that the student model can perform inference operations based on the second sample data to calculate the first probability distribution of different element data. Specifically, this can be implemented as follows:

[0125] The second sample data is input into the student model and the teacher model respectively, so that the student model can perform inference operations based on the second sample data and select target element data according to the first probability distribution.

[0126] In some embodiments, the method may further include:

[0127] The target element data is concatenated with the second sample data to update the second sample data, and the process returns to the step of inputting the second sample data into the student model and the teacher model respectively to continue execution.

[0128] In the embodiments of this application, after the splicing is completed, the updated second sample data is input again into the student model and the teacher model, and the above steps are repeated to realize an iterative training process. By continuously updating the sample data and letting the model process it, the model can gradually learn how to generate more reasonable and coherent text or dialogue, and can continuously optimize its calculation of the probability distribution of element data and the selection of target element data, which helps to improve the performance of the model and make it perform better in tasks such as dialogue generation and text continuation.

[0129] In practical applications, this iterative training process can continue multiple times until certain stopping conditions are met. These stopping conditions may include reaching a predetermined number of iterations, achieving a required text length, or the generated content meeting certain evaluation metrics (such as generating dialogue that conforms to human language habits and logic). Through this iterative approach, the student model can continuously learn from the input data and gradually adjust its parameters and predictive capabilities, ultimately learning and absorbing knowledge from the teacher model to achieve better performance. Simultaneously, the teacher model can also provide a reference for the student model, helping it adjust its predictions of element data in each iteration, gradually bringing the student model closer to the performance of the teacher model.

[0130] Figure 3 The flowchart shown is an example of an artificial intelligence interview method provided in one embodiment of this application, applied to a server, such as... Figure 3 As shown, the method may specifically include the following steps:

[0131] 301: Retrieved the interview request sent by the client; the interview request is generated by the client in response to the interview action triggered by the target user for the target position.

[0132] In the embodiments of this application, the server can detect interview actions triggered by the client. When a target user triggers an interview action for a specific position on the display interface provided by the client, the client can generate an interview request. The interview action could be, for example, the target user clicking the "Start Interview" button or selecting to apply for an interview on a specific recruitment page.

[0133] The client can then send this interview request to the server. Upon receiving the request, the server can initiate the interview process. For example, on an online recruitment platform, after a user browses a desired waiter / waitress position and clicks the "Apply for Interview" button, the client will generate the corresponding interview request and send it to the server.

[0134] 302: Determine the question list corresponding to the target position; the question list includes multiple question data.

[0135] After receiving an interview request, the server can retrieve a series of questions related to the target position. The server can use its internal database or configuration information to find the question list corresponding to the target position. This question list contains multiple sets of questions, which may be pre-prepared based on the characteristics, requirements, and common assessment points of the target position. For example, for a software engineer position, the question list might include questions about programming language knowledge, project experience, and algorithm understanding.

[0136] 303: Determine the target dialogue model; the target dialogue model is obtained by transferring knowledge from the second dialogue model to the first dialogue model; the first dialogue model is used as the student model, and the second dialogue model is used as the teacher model; the student model is adjusted based on the difference information of the first probability distribution and the second probability distribution of the different element data generated by reasoning; the first probability distribution is generated by the student model based on the second sample data; the second probability distribution is generated by the teacher model based on the second sample data; the second sample input model is obtained by concatenating the first sample data with the target element data; the target element data is generated by the student model based on the first sample data.

[0137] The target dialogue model can be achieved through Figure 2 The dialogue model shown was trained using a specific training method. The training method for the target dialogue model can be referenced from this example. Figure 2 The relevant embodiments of the dialogue model training method shown will not be described again here.

[0138] 304: Generate target question data from the question list by using the target dialogue model in conjunction with at least one round of historical dialogue data.

[0139] The target dialogue model can generate target question data from a question list based on at least one round of historical dialogue data. This historical dialogue data records the content already discussed during previous interviews. The target dialogue model can analyze this historical dialogue data and the question list to filter and generate target question data most suitable for the current interview stage.

[0140] The question list contains multiple pre-defined questions, which can be arranged in a specific logical order. The target dialogue model can be trained to generate target question data from the question list, following a predefined question order.

[0141] The target dialogue model can determine which question should be output as the target question data based on the input historical dialogue data, its own training results, and the order of questions from the question list.

[0142] 305: Sends the target problem data to the client.

[0143] 306: Obtain the response data from the client and return to the step of generating target question data from the question list using the target dialogue model combined with at least one round of historical dialogue data to continue execution.

[0144] After generating the target question data, the server can send it to the client. Upon receiving the target question data, the client can display it to the target user, allowing the user to answer. This way, the target user can see the interview questions and prepare their responses.

[0145] The client can collect the target user's responses to the target question and send them back to the server. After receiving this response data, the server returns to step 304, where it uses the target dialogue model combined with at least one new round of historical dialogue data (which includes the user's responses) to generate new target question data from the question list. This process loops continuously until the interview ends, thus creating a dynamic interview process that progresses based on user responses.

[0146] In some embodiments, selecting target question data from a question list using a target dialogue model combined with at least one round of historical dialogue data can be specifically implemented as follows:

[0147] Based on at least one round of historical dialogue data, determine whether the target user meets the question generation conditions;

[0148] If the target user meets the question generation conditions, target question data is generated from the question list using the target dialogue model and at least one round of historical dialogue data.

[0149] The historical dialogue data from at least one round includes information about previous interview interactions. This information can help the target dialogue model determine whether new question data needs to be generated. If new question data needs to be generated, the target dialogue model can be used, combined with the historical dialogue data from at least one round, to generate target question data from the question list.

[0150] In some embodiments, if the target user does not meet the problem generation conditions, the method may further include:

[0151] Identify target question data from historical question data in historical dialogue data.

[0152] If the target user does not meet the conditions for generating questions, that is, no new question data needs to be generated at present, the target question data can be determined from the historical question data that has already been asked.

[0153] In some embodiments, determining whether a target user meets the question generation conditions by combining at least one round of historical dialogue data can be specifically implemented as follows:

[0154] Identify the previous round's question data and response data from the previous round of dialogue;

[0155] Determine if the previous round of response data matches the previous round of question data; if so, determine if the target user meets the question generation conditions; otherwise, the question generation conditions are not met.

[0156] In one possible implementation, a first recognition model can be used to determine whether the response data from the previous round matches the question data from the previous round. The first recognition model can be a trained large model. A large model refers to a machine learning model with a large number of parameters and a complex structure, capable of processing massive amounts of data and completing various complex tasks, such as natural language processing, computer vision, and speech recognition; it is a type of AI (Artificial Intelligence) model. For example, a large model can be implemented using a Large Language Model (LLM) or a Multimodal Large Model (MLM), such as GPT-3 (Generative Pre-Trained Transformer-3), GPT-4 (Generative Pre-Trained Transformer-4), BERT (Bidirectional Encoder Representation from Transformers), Turing NLG (Turing Natural Language Generation), etc. This application does not limit this approach. These large models can perform well in a variety of natural language processing tasks.

[0157] The first recognition model can be trained based on sample dialogue data and the matching labels of sample question and response data within the sample dialogue data. For example, sample dialogue data may contain multiple sets of sample question and response data, as well as corresponding matching labels (indicating whether the response matches the question), such as "Question: Please list the programming languages ​​you are proficient in. Response: I am proficient in Java and Python. Matching label: Yes." By training on a large amount of such sample data, the first recognition model learns how to determine whether a response reasonably answers the corresponding question.

[0158] When the question and response data from the previous round are received, they can be input into the first recognition model. The first recognition model can then use its trained knowledge to evaluate whether the response data from the previous round is a reasonable answer to the question data from the previous round.

[0159] If the first identification model determines that the previous round of response data matches the previous round of question data, it means that more questions need to be asked to gain a deeper understanding of the user's situation, or to proceed to the next interview stage. This indicates that the target user meets the conditions for question generation.

[0160] If the first identification model determines that the previous round of response data does not match the previous round of question data, it may mean that the user's answer is not accurate enough or that the user does not understand the question. It is necessary to ask the current question again. At this time, it can be determined that the target user does not meet the question generation conditions.

[0161] In some embodiments, the method may further include:

[0162] If the target user does not meet the conditions for generating the issue, the issue data from the previous round will be used as the target issue data.

[0163] In some embodiments, if the target user does not meet the problem generation conditions, using the previous round of problem data as the target problem data can be specifically implemented as follows:

[0164] If the target user does not meet the conditions for generating the problem, determine whether the number of retries for the previous round of problem data meets the limit.

[0165] If not, use the previous round of problem data as the target problem data and accumulate the number of retries;

[0166] If so, use the target dialogue model and combine it with at least one round of historical dialogue data to generate target question data from the question list.

[0167] The retry count refers to the number of times the same question is asked to the target user. The limitation can be a pre-set threshold to control the maximum number of times the same question can be asked. For example, in an interview scenario, a limitation could be set that each question can be retried a maximum of 3 times. If this is the first or second time asking the target user a question, it's necessary to determine if the current retry count has reached 3. If it hasn't reached 3, the limitation is not met, and the question can be asked again; if it has reached 3, the limitation is met.

[0168] If it is determined that the retries for the previous round of question data have not met the limit, then the previous round of question data will be sent to the target user again as the target question data. If the previous round of response data does not match the previous round of question data, it may be that the target user did not understand the question or gave an incorrect answer. By asking again, a more appropriate answer may be obtained.

[0169] At the same time, after sending out the previous round of problem data as the target problem data, the number of retries for that problem needs to be accumulated.

[0170] When it's determined that the retries for the previous round of questions have met the limit, it means that the target user still hasn't provided a suitable answer after multiple attempts. At this point, to prevent the interview or conversation from getting stuck on this question, it can be temporarily skipped to ensure the smooth progress of the interview or conversation.

[0171] In some embodiments, if the target user does not meet the problem generation conditions, using the previous round of problem data as the target problem data can be specifically implemented as follows:

[0172] If the target user does not meet the question generation conditions, determine whether the previous round of response data is question data;

[0173] If not, use the problem data from the previous round as the target problem data;

[0174] If so, use the conversation model to generate the response data corresponding to the question data, and send the response data to the client.

[0175] In one possible implementation, a second identification model can be used to determine whether the previous round of response data was question data. The second identification model can be a large, trained model.

[0176] If the second identification model determines that the previous round of response data is not question data, it means that the user is answering a question, but their answer may not meet the requirements, causing the target user to not meet the question generation conditions. In this case, the previous round of question data can be used as the target question data.

[0177] When the second recognition model determines that the previous round of response data was question data, it means that the target user has raised a question. At this point, a conversation model is needed to generate response data for this question data based on the job information of the target position.

[0178] In one possible implementation, the conversation model can be implemented as a target dialogue model.

[0179] In some embodiments, determining whether a target user meets the question generation conditions by combining at least one round of historical dialogue data can be specifically implemented as follows:

[0180] Identify the previous round's question data and response data from the previous round of dialogue;

[0181] Determine if the previous round of response data was question data; if so, determine if the target user meets the question generation conditions.

[0182] In some embodiments, the method may further include:

[0183] If the target user does not meet the conditions for generating the issue, the issue data from the previous round will be used as the target issue data.

[0184] In some embodiments, if the historical dialogue data meets the question generation conditions, generating target question data from the question list using the target dialogue model and combining it with at least one round of historical dialogue data can be specifically implemented as follows:

[0185] If the target user meets the question generation conditions, the conversation model is used to generate the corresponding response data for the question data, and the response data is sent to the client.

[0186] Target question data is generated from a question list using a target dialogue model and historical dialogue data.

[0187] In some embodiments, sending the target problem data to the client can be specifically implemented as follows:

[0188] Determine whether the target problem data meets the constraints;

[0189] If so, send the target problem data to the client;

[0190] If not, return to the previous step and use the target dialogue model in conjunction with at least one round of historical dialogue data to generate target question data from the question list.

[0191] In some embodiments, determining whether the target problem data satisfies the constraints can be specifically implemented as follows:

[0192] Determine whether at least one round of historical dialogue data contains prior information about the target question;

[0193] If yes, then the constraint conditions are satisfied; otherwise, the constraint conditions are not satisfied.

[0194] In this context, at least one question in the question list can be configured with constraints. A question with constraints means that its formulation requires certain prior information. If this prior information appears in at least one round of historical dialogue data, the target question can be considered to satisfy the constraints; otherwise, it does not. For example, in an interview scenario, for the question "Please elaborate on the specific problems you encountered and their solutions when using Python for data analysis," the prior information could be that the target user mentioned having experience using Python in a previous conversation.

[0195] After selecting target question data with configured constraints, the target dialogue model can examine at least one round of historical dialogue data to see if there is any pre-existing information related to the target question data. If so, it indicates that the target question data meets the constraints because it is a natural extension of the previous dialogue content and conforms to the logical coherence of the dialogue. For example, if the job applicant mentioned in a previous dialogue that "I often use Python to process data at work," then the aforementioned question about Python data analysis has a reasonable pre-existing condition and meets the constraints.

[0196] If the target question data meets the constraints, the server can send it to the client. This ensures that the question received by the client is relevant and reasonable to previous conversations, facilitating smooth dialogue progression. If the target question data does not meet the constraints, the server may return to the previous state and perform an operation to generate the target question data from the question list using the target dialogue model and at least one round of historical dialogue data.

[0197] In one possible implementation, a third identification model can be used to determine whether the target problem data satisfies the constraints. The third identification model can be a large, trained model.

[0198] In real-world applications, different groups of people, with varying ages, occupations, educational backgrounds, and experiences, will require different questions in conversations. For instance, in job interviews, interviewers focus on different aspects than recent graduates and experienced candidates, leading to different questions. Furthermore, the specific questions asked will vary depending on the job opening.

[0199] By defining pre-defined constraints for the question data, it's like setting a "switch" for the questions. Only when specific conditions are met will the corresponding question be asked. When expanding to new categories, such as adding new job categories in recruitment, you only need to define all the questions to be asked for that category, and clarify the conditions under which different questions will be asked. The target dialogue model can then automatically raise appropriate questions in the new category's dialogue scenario based on these settings, thereby achieving automated support for the new category's dialogue capabilities. This eliminates the need for large-scale modifications to the overall dialogue system's architecture or logic, reducing development and maintenance costs and improving the flexibility and scalability of interviews.

[0200] In some embodiments, generating target question data from a question list using a target dialogue model combined with at least one round of historical dialogue data can be specifically implemented as follows:

[0201] Using the target dialogue model combined with at least one round of historical dialogue data, candidate question data is generated from the unmarked question data in the question list. Then, it is determined whether there is corresponding response data for the candidate question data from the at least one round of historical dialogue data. If so, the candidate question data is marked and the candidate question data is regenerated. If not, the candidate question data is used as the target question data and the target question data is marked.

[0202] Unlabeled question data refers to questions that have not been used in previous conversations. The target dialogue model can generate candidate question data from this unlabeled question data, based on information provided by historical dialogue data.

[0203] If a question already has a relevant answer, then there may be no need to ask the question again. Therefore, when generating candidate question data, search at least one round of historical dialogue data to see if there is already a response to the candidate question.

[0204] If a response to a candidate question is found in the historical dialogue data, that candidate question is marked. Marking the question indicates that the question already has relevant information in the current dialogue and cannot be considered a valid question. Then, candidate question data may be regenerated, and a suitable question may be searched from the unmarked questions in the question list, repeating the above judgment process.

[0205] If no response data corresponding to a candidate question is found in the historical dialogue data, it means that this question has not been discussed in the current dialogue and is a new question that can be asked. In this case, the candidate question data can be used as the target question data and marked. Marking the target question data is to record that this question has been used and will not be mistakenly selected again in subsequent dialogues.

[0206] In some embodiments, generating target question data from a question list using a target dialogue model combined with at least one round of historical dialogue data can be specifically implemented as follows:

[0207] Using the target dialogue model and at least one round of historical dialogue data, generate target question data from the question list according to at least one question output requirement;

[0208] At least one question output requirement includes one or more of the following options:

[0209] Matching response data with question data from the previous round of historical dialogue data;

[0210] The number of retries for the question data in the previous round of historical dialogue data exceeded the limit;

[0211] There is no response data matching the target question in at least one round of historical dialogue data;

[0212] as well as,

[0213] The target problem data satisfies its corresponding constraints.

[0214] As mentioned in the above embodiments, in this application embodiment, a first recognition model, a second recognition model, and a third recognition model can be used to identify whether the response data matches the question data, whether the number of retries exceeds the limit, and whether the constraints are met. This allows the complex tasks that the target dialogue model needs to handle to be broken down, thereby reducing the processing pressure on the target dialogue model and lowering its training cost. Furthermore, each recognition model can be trained using more targeted data, allowing different recognition models to focus more on their respective tasks. Through specialized training and optimization, their performance on specific tasks can be improved.

[0215] In some embodiments, after obtaining the response data from the client, the method may further include:

[0216] Determine if there are any unselected question data in the question list;

[0217] If yes, select the question data step from the question list and continue; otherwise, generate an interview end message.

[0218] If, after checking the question list, no unselected questions are found, meaning all questions have been asked, this indicates the interview has covered all pre-defined assessment aspects, and the interview can end. At this point, the server can generate an interview completion message. This message is sent to the client to inform the applicant that the interview is over. For example, the client might display "This interview has ended. Thank you for your participation!"

[0219] In some embodiments, the method may further include:

[0220] Receive resume submissions from clients for the target position;

[0221] In response to a resume submission, send an interview invitation notification to the client.

[0222] The interview request sent by the client includes:

[0223] Retrieve the interview request sent by the client when it triggers an interview invitation notification.

[0224] Figure 4 A schematic diagram of the artificial intelligence interview method provided in an embodiment of this application is shown.

[0225] exist Figure 4 In this context, 401 can represent the server and 402 can represent the client. The target dialogue model 4011 can be deployed on the server 401, and the server 401 can deploy the first recognition model 4012, the second recognition model 4013, and the third recognition model 4014.

[0226] After receiving the response data for the target question data sent by the client 402, the server 401 can input at least one round of historical dialogue data into the first recognition model 4012, the second recognition model 4013, and the third recognition model 4014 respectively. The first recognition model 4012 can identify whether the target user's response data matches the target question data and whether there is a question in the target user's response data; the second recognition model 4013 can be used to identify the target question data in the current round of question and answer and the number of retries for the target question data; and the third recognition model 4014 can be used to identify whether the target question data meets the constraints.

[0227] The first recognition model 4012, the second recognition model 4013, and the third recognition model 4014 can each output recognition results. These results can then be input into the target dialogue model, allowing it to generate question data for the next round of dialogue based on the recognition results. The question data for the next round of dialogue can be either a repetition of the target question data or a reselection of question data from the question list.

[0228] In the embodiments of this application, to improve the interactive experience, after determining the target question data, the server can convert the target question data into voice data, so that a digital human can verbally broadcast the target question data using the generated voice data. Using a digital human for verbal broadcasting creates an atmosphere closer to real face-to-face communication for the target user.

[0229] Figure 5 This application provides a block diagram of a training apparatus for a dialogue model according to one embodiment. Figure 5 As shown, the device may include:

[0230] The first determining module 501 is used to determine a first dialogue model and a second dialogue model, wherein the number of parameters of the second dialogue model is greater than the number of parameters of the first dialogue model.

[0231] The second determining module 502 is used to use the first dialogue model as a student model and the second dialogue model as a teacher model.

[0232] The sample acquisition module 503 is used to acquire first sample data; the first sample data includes at least one round of sample dialogue data.

[0233] The first input module 504 is used to input the first sample data into the student model, so as to use the student model to perform inference operations based on the first sample data, to calculate the first probability distribution of different element data, and select target element data according to the first probability distribution; the element data is the smallest data unit.

[0234] The splicing module 505 is used to splice the target element data with the first sample data to obtain the second sample data;

[0235] The second input module 506 is used to input the second sample data into the student model and the teacher model respectively, so as to use the student model to perform inference operations based on the second sample data to calculate the first probability distribution of different element data, and use the teacher model to perform inference operations based on the second sample data to calculate the second probability distribution of different element data.

[0236] The difference determination module 507 is used to determine the difference information between the first probability distribution and the second probability distribution corresponding to different element data;

[0237] Training module 508 is used to adjust the student model based on the difference information to transfer the knowledge of the second dialogue model to the first dialogue model and obtain a target dialogue model; the target dialogue model is used to generate target question data from a list of questions corresponding to the target position by combining at least one round of historical dialogue data.

[0238] In some embodiments, the device may further include:

[0239] A pre-training data acquisition module is used to acquire pre-training data, which includes third sample data and sample question data corresponding to the third sample data; the third sample data includes at least one round of sample dialogue data; each round of sample dialogue data includes sample question data and sample response data corresponding to the question data.

[0240] The pre-training module is used to pre-train the first dialogue model and the second dialogue model using the pre-training data.

[0241] In some embodiments, the pre-training module is specifically used for:

[0242] For either the first dialogue model or the second dialogue model, the third sample data is used as the input data of the model, and the sample question data is used as the training label.

[0243] The model is trained using the input data and the training labels.

[0244] In some embodiments, the sample acquisition module 503 is specifically used for:

[0245] Determine at least one round of sample dialogue data, and the required question output;

[0246] The first sample data is generated based on the at least one round of sample dialogue data and the question output requirements.

[0247] In some embodiments, the second input module 506 is specifically used for:

[0248] The second sample data is input into the student model and the teacher model respectively, so that the student model can perform inference operations based on the second sample data and select target element data according to the first probability distribution.

[0249] In some embodiments, the device further includes:

[0250] The data splicing module is used to splice the target element data with the second sample data to update the second sample data, and then return to the step of inputting the second sample data into the student model and the teacher model respectively to continue execution.

[0251] Figure 5 The training device for the dialogue model can perform... Figure 2 The implementation principle and technical effects of the dialogue model method described in the illustrated embodiments will not be repeated here. The specific methods by which each module and unit of the dialogue model apparatus in the above embodiments perform operations have been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0252] Figure 6 The diagram shows a block diagram of an artificial intelligence interview device provided in an embodiment of this application. This device can be applied to a server, such as... Figure 6 As shown, the device may include:

[0253] The first request acquisition module 601 is used to acquire the interview request sent by the client; the interview request is generated by the client in response to the interview operation triggered by the target user for the target position;

[0254] The first question list determination module 602 is used to determine the question list corresponding to the target position; the question list includes multiple question data.

[0255] A dialogue model determination module 603 is used to determine a target dialogue model; the target dialogue model is obtained by transferring knowledge from a second dialogue model to a first dialogue model; the first dialogue model is used as a student model, and the second dialogue model is used as a teacher model; the student model is adjusted based on the difference information between a first probability distribution and a second probability distribution of different element data generated through reasoning; the first probability distribution is generated by the student model based on second sample data; the second probability distribution is generated by the teacher model based on second sample data; the second sample input model is obtained by concatenating the first sample data with the target element data; the target element data is generated by the student model based on the first sample data.

[0256] The first question generation module 604 is used to generate target question data from the question list by combining the target dialogue model with at least one round of historical dialogue data.

[0257] The first question sending module 605 is used to send the target question data to the client;

[0258] The first response acquisition module 606 is used to acquire the response data fed back by the client and return to the step of generating target question data from the question list by combining the target dialogue model with at least one round of historical dialogue data to continue execution.

[0259] In some embodiments, the first problem generation module 604 is specifically used for:

[0260] Based on at least one round of historical dialogue data, determine whether the target user meets the question generation conditions;

[0261] If the target user meets the question generation conditions, target question data is generated from the question list using the target dialogue model and in combination with at least one round of historical dialogue data.

[0262] In some embodiments, if the target user does not meet the problem generation conditions, the device may further include:

[0263] The first question determination module is used to determine the target question data from the historical question data in the historical dialogue data.

[0264] In some embodiments, the first problem generation module 604 is specifically used for:

[0265] Identify the previous round's question data and response data from the previous round of dialogue;

[0266] Determine whether the previous round of response data matches the previous round of question data; if so, determine that the target user meets the question generation conditions; otherwise, the question generation conditions are not met.

[0267] In some embodiments, the device may further include:

[0268] The second problem determination module is used to use the problem data from the previous round as the target problem data when the target user does not meet the problem generation conditions.

[0269] In some embodiments, the second problem-determining module is specifically used for:

[0270] If the target user does not meet the conditions for generating the problem, determine whether the number of retries for the previous round of problem data meets the limit.

[0271] If not, use the previous round of problem data as the target problem data and accumulate the number of retries;

[0272] If so, using the target dialogue model and combining it with at least one round of historical dialogue data, target question data is generated from the question list.

[0273] In some embodiments, the second problem-determining module is specifically used for:

[0274] If the target user does not meet the question generation conditions, determine whether the previous round of response data is question data;

[0275] If not, use the problem data from the previous round as the target problem data;

[0276] If so, the response data corresponding to the question data is generated using the conversation model, and the response data is sent to the client.

[0277] In some embodiments, the second problem-determining module is specifically used for:

[0278] Identify the previous round's question data and response data from the previous round of dialogue;

[0279] Determine whether the previous round of response data is question data; if so, determine that the target user meets the question generation conditions.

[0280] In some embodiments, the device further includes:

[0281] The third problem determination module is used to use the previous round of problem data as the target problem data if the target user does not meet the problem generation conditions.

[0282] In some embodiments, the first problem generation module 604 is specifically used for:

[0283] If the target user meets the question generation conditions, the response data corresponding to the question data is generated using the conversation model, and the response data is sent to the client.

[0284] Target question data is generated from the question list using the target dialogue model and the historical dialogue data.

[0285] In some embodiments, the first question sending module 605 is specifically used for:

[0286] Determine whether the target problem data meets the constraints;

[0287] If so, the target problem data is sent to the client;

[0288] If not, return to the process of generating target question data from the question list by combining the target dialogue model with at least one round of historical dialogue data.

[0289] In some embodiments, the first question sending module 605 is specifically used for:

[0290] Determine whether the at least one round of historical dialogue data contains prior information about the target question data;

[0291] If yes, then the constraint condition is satisfied; otherwise, the constraint condition is not satisfied.

[0292] In some embodiments, the first problem generation module 604 is specifically used for:

[0293] Using the target dialogue model combined with at least one round of historical dialogue data, candidate question data is generated from the unmarked question data in the question list. Then, it is determined whether there is response data corresponding to the candidate question data in the at least one round of historical dialogue data. If so, the candidate question data is marked and the candidate question data is regenerated. If not, the candidate question data is used as the target question data and the target question data is marked.

[0294] In some embodiments, the first problem generation module 604 is specifically used for:

[0295] Using the target dialogue model and at least one round of historical dialogue data, target question data is generated from the question list according to at least one question output requirement;

[0296] The output requirement for at least one question includes one or more of the following options:

[0297] Matching response data with question data from the previous round of historical dialogue data;

[0298] The number of retries for the question data in the previous round of historical dialogue data exceeded the limit;

[0299] There is no response data matching the target question in at least one round of historical dialogue data;

[0300] as well as,

[0301] The target problem data satisfies its corresponding constraints.

[0302] In some embodiments, the device may further include:

[0303] The list determination module is used to determine whether there is any unselected question data in the question list;

[0304] The first selection module is used to select question data from the question list if there is unselected question data in the question list and continue the process; otherwise, it generates an interview end prompt message.

[0305] In some embodiments, the device may further include:

[0306] The first operation receiving module is used to receive the resume submission operation of the client for the target position;

[0307] The first notification module is used to send an interview invitation notification to the client in response to the resume submission operation;

[0308] In some embodiments, the first request acquisition module 601 is specifically used for:

[0309] Obtain the interview request sent by the client by triggering the interview invitation notification.

[0310] Figure 6 The aforementioned AI interview device can perform Figure 3 The implementation principle and technical effects of the AI ​​interview method described in the illustrated embodiments will not be repeated here. The specific methods by which each module and unit of the AI ​​interview device in the above embodiments performs its operations have been described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0311] This application also provides a computing device, such as... Figure 7 As shown, the device may include a storage component and a processing component;

[0312] The storage component contains one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component to implement the dialogue model training method and artificial intelligence interview method provided in the embodiments of this application.

[0313] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0314] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc. Communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0315] The processing component may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0316] Storage components are configured to store various types of data to support operations on the terminal. Storage components can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0317] The display component can be an electroluminescent (EL) element, a liquid crystal display or a microdisplay with a similar structure, or a retina-direct display or a similar laser scanning display.

[0318] It should be noted that the aforementioned computing devices can be physical devices or elastic computing hosts provided by cloud computing platforms. They can be implemented as a distributed cluster of multiple servers or terminal devices, or as a single server or a single terminal device.

[0319] This application also provides a computer-readable storage medium storing a computer program. When executed by a computer, the computer program can implement the training method for the dialogue model and the artificial intelligence interview method provided in the above embodiments. This computer-readable medium may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device.

[0320] This application also provides a computer program product, which includes a computer program carried on a computer-readable storage medium. When executed by a computer, the computer program can implement the training method for the dialogue model and the artificial intelligence interview method provided in the above embodiments. In such embodiments, the computer program may be downloaded and installed from a network, and / or installed from a removable medium. When the computer program is executed by a processor, it performs the various functions defined in the system of this application.

[0321] It should be noted that the embodiments of this application may involve the use of user data. In practical applications, user-specific personal data may be used in the scheme described herein within the scope permitted by applicable laws and regulations, provided that it complies with the applicable laws and regulations of the country (e.g., with the user's explicit consent, with the user being properly notified, etc.).

[0322] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0323] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0324] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0325] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for training a dialogue model, the method comprising: The method comprises: determining a first dialogue model and a second dialogue model, wherein a parameter quantity of the second dialogue model is greater than a parameter quantity of the first dialogue model; taking the first dialogue model as a student model and the second dialogue model as a teacher model; obtaining first sample data; the first sample data comprises at least one round of sample dialogue data; inputting the first sample data into the student model to perform an inference operation based on the first sample data by using the student model, to calculate a first probability distribution of different element data, and to select target element data according to the first probability distribution; the element data is a minimum data unit; splicing the target element data and the first sample data to obtain second sample data; inputting the second sample data into the student model and the teacher model respectively, to perform an inference operation based on the second sample data by using the student model, to calculate a first probability distribution of different element data, and to perform an inference operation based on the second sample data by using the teacher model, to calculate a second probability distribution of different element data; determining difference information of the first probability distribution and the second probability distribution corresponding to different element data respectively; adjusting the student model based on the difference information, to transfer the knowledge of the second dialogue model to the first dialogue model, to obtain a target dialogue model; the target dialogue model is used to generate target question data from a question list corresponding to a target position in combination with at least one round of historical dialogue data.

2. The method of claim 1, wherein, The method further comprises: obtaining pre-training data; the pre-training data comprises third sample data and sample question data corresponding to the third sample data; the third sample data comprises at least one round of sample dialogue data; each round of sample dialogue data comprises sample question data and sample answer data corresponding to the question data; pre-training the first dialogue model and the second dialogue model respectively by using the pre-training data.

3. The method of claim 2, wherein, The pre-training the first dialogue model and the second dialogue model respectively by using the pre-training data comprises: for any one of the first dialogue model and the second dialogue model, taking the third sample data as input data of the model, and taking the sample question data as a training label; training the model by using the input data and the training label.

4. The method of claim 1, wherein, The obtaining first sample data comprises: determining at least one round of sample dialogue data and question output requirements; generating the first sample data based on the at least one round of sample dialogue data and the question output requirements.

5. The method of claim 1, wherein, The inputting the second sample data into the student model and the teacher model respectively, to perform an inference operation based on the second sample data by using the student model, to calculate a first probability distribution of different element data, and to select target element data according to the first probability distribution comprises: inputting the second sample data into the student model and the teacher model respectively, to perform an inference operation based on the second sample data by using the student model, and to select target element data according to the first probability distribution; The method further comprises: The target element data is spliced with the second sample data to update the second sample data, and the step of inputting the second sample data into the student model and the teacher model respectively is continued to be executed.

6. An artificial intelligence interview method, characterized by, The method applied to a server comprises: obtaining an interview request sent by a client; the interview request is generated by the client in response to an interview operation triggered by a target user on a target position; determining a question list corresponding to the target position; the question list comprises a plurality of question data; determining a target dialogue model; the target dialogue model is obtained by migrating the knowledge of a second dialogue model to a first dialogue model; the first dialogue model is used as a student model, and the second dialogue model is used as a teacher model; the student model is used to adjust based on difference information of a first probability distribution and a second probability distribution of different element data generated by reasoning; the first probability distribution is generated by the student model based on second sample data; the second probability distribution is generated by the teacher model based on the second sample data; the second sample input model is obtained by splicing target element data from first sample data; the target element data is generated by the student model based on the first sample data; generating target question data from the question list by using the target dialogue model in combination with at least one round of historical dialogue data; sending the target question data to the client; obtaining response data fed back by the client, and continuing to execute the step of generating target question data from the question list by using the target dialogue model in combination with at least one round of historical dialogue data.

7. The method of claim 6, wherein, The step of generating target question data from the question list by using the target dialogue model in combination with at least one round of historical dialogue data comprises: determining whether the target user meets a question generation condition in combination with at least one round of historical dialogue data; if the target user meets the question generation condition, generating target question data from the question list by using the target dialogue model in combination with at least one round of historical dialogue data.

8. The method of claim 7, wherein, If the target user does not meet the question generation condition, the method further comprises: determining target question data from historical question data in the historical dialogue data.

9. The method of claim 8, wherein, The step of determining whether the target user meets the question generation condition in combination with at least one round of historical dialogue data comprises: determining previous round question data and previous round response data in previous round dialogue data; determining whether the previous round response data matches the previous round question data; if yes, determining that the target user meets the question generation condition, otherwise, determining that the target user does not meet the question generation condition; The method further comprises: if the target user does not meet the question generation condition, using the previous round question data as the target question data.

10. The method of claim 9, wherein, The step of using the previous round question data as the target question data if the target user does not meet the question generation condition comprises: if the target user does not meet the question generation condition, determining whether a retry number of the previous round question data meets a limit condition; if no, using the previous round question data as the target question data, and accumulating the retry number; If yes, the target question data is generated from the question list by using the target dialogue model in combination with at least one round of historical dialogue data.

11. The method of claim 9, wherein, If the target user does not meet the question generation condition, the previous round of question data is included as the target question data. If the target user does not meet the question generation condition, it is determined whether the previous round of response data is question data. If no, the previous round of question data is included as the target question data. If yes, the reply data corresponding to the question data is generated by using the dialogue model, and the reply data is sent to the client.

12. The method of claim 7, wherein, The combination of at least one round of historical dialogue data to determine whether the target user meets the question generation condition includes: Determine the previous round of question data and the previous round of response data in the previous round of dialogue data. If yes, it is determined that the target user meets the question generation condition. The method further includes: If the target user does not meet the question generation condition, the previous round of question data is included as the target question data. If the target user meets the question generation condition, the reply data corresponding to the question data is generated by using the dialogue model, and the reply data is sent to the client. The target question data is generated from the question list by using the target dialogue model in combination with the historical dialogue data. The target question data is sent to the client, which includes:

13. The method of claim 6, wherein, Determine whether the target question data meets the constraint condition. If yes, the target question data is sent to the client. If no, return to execute the target question data generated from the question list by using the target dialogue model in combination with at least one round of historical dialogue data. The determination of whether the target question data meets the constraint condition includes:

14. The method of claim 13, wherein, Determine whether the at least one round of historical dialogue data contains the prerequisite information of the target question data. If yes, it is determined that the constraint condition is met, and if no, it is determined that the constraint condition is not met. The target question data is generated from the question list by using the target dialogue model in combination with at least one round of historical dialogue data, which includes:

15. The method of claim 6, wherein, The candidate question data is generated from the unlabeled question data in the question list by using the target dialogue model in combination with at least one round of historical dialogue data, and it is determined whether the response data corresponding to the candidate question data exists in the at least one round of historical dialogue data. If yes, the candidate question data is labeled, and the candidate question data is regenerated. If no, the candidate question data is included as the target question data, and the target question data is labeled. The target question data is generated from the question list by using the target dialogue model in combination with at least one round of historical dialogue data, which includes:

16. The method of claim 6, wherein, The target question data is generated from the question list by using the target dialogue model in combination with at least one round of historical dialogue data according to at least one question output requirement. The at least one question output requirement includes one or more options as follows: The response data in the previous round of historical dialogue data matches the question data. ​ The number of retries of the question data in the previous round of historical dialogue data exceeds the limit condition; There is no answer data matching the target question data in at least one round of historical dialogue data; And, The target question data meets its corresponding constraint condition.

17. The method of claim 6, wherein, After obtaining the answer data of the client feedback, the method further comprises: Determining whether there is unselected question data in the question list; If yes, continue to execute the question data selection step from the question list, and if no, generate an interview end prompt information.

18. The method of claim 6, wherein, Further comprising: Receiving a resume submission operation of the client for the target position; In response to the resume submission operation, sending an interview invitation notice to the client; The interview request sent by the client comprises: Obtaining the interview request sent by the client by triggering the interview invitation notice.

19. A computing device, comprising: Comprise a processing component and a storage component; The storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to realize the training method of the dialogue model as claimed in any one of claims 1 to 5, or realize the artificial intelligence interview method as claimed in any one of claims 6 to 18.

20. A computer-readable storage medium, characterized in that, The computer program stored thereon is executed by the processing component to realize the training method of the dialogue model as claimed in any one of claims 1 to 5, or realize the artificial intelligence interview method as claimed in any one of claims 6 to 18.

21. A computer program product, characterised in that, Comprise computer program / instructions, which are executed by the processing component to realize the training method of the dialogue model as claimed in any one of claims 1 to 5, or realize the artificial intelligence interview method as claimed in any one of claims 6 to 18.

Images