Simulation interview method, device and product based on large model

Abstract

The invention provides a simulation interview method and device based on a large model, electronic equipment, a storage medium and a computer program product, relates to the technical field of artificial intelligence, in particular to the technical fields of large models, natural language understanding, simulation interview and the like, and can be applied to simulation interview scenes. According to the specific implementation scheme, through an artificial intelligence large model, an interview question for an interview object is generated according to historical question and answer data of the interview object in the interview process; obtaining response data of the interview object to the interview question; and determining the response quality of the interview object to the interview question according to the response data through an artificial intelligence large model. According to the invention, the logicality of interview problems in the interview process is improved, and the interactivity between the interview process and the user is simulated; the response quality is immediately determined according to the response data of the interview question, and the simulation interview requirement of the interview object is met.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of artificial intelligence, in particular to the technical fields of artificial intelligence large model, natural language understanding, simulated interview, and the like, and more particularly to a simulated interview method and device based on a large model, an electronic device, a storage medium, and a computer program product, which can be applied to a simulated interview scenario. BACKGROUND

[0002] The current simulated interview system usually presets an interview question library, and determines a target question from the interview question library according to a fixed order or randomly to ask the interviewee. The system judges the answer quality of the interviewee based on a keyword matching algorithm by recognizing the voice of the interviewee. SUMMARY

[0003] The present disclosure provides a simulated interview method and device based on a large model, an electronic device, a storage medium, and a computer program product.

[0004] According to a first aspect, a simulated interview method based on a large model is provided, including: generating, by an artificial intelligence large model, an interview question for an interviewee according to historical question and answer data in an interview process of the interviewee; obtaining answer data of the interviewee to the interview question; and determining, by the artificial intelligence large model, an answer quality of the interviewee to the interview question according to the answer data.

[0005] According to a second aspect, a simulated interview device based on a large model is provided, including: a question generation unit configured to generate, by an artificial intelligence large model, an interview question for an interviewee according to historical question and answer data in an interview process of the interviewee; an answer obtaining unit configured to obtain answer data of the interviewee to the interview question; and a quality evaluation unit configured to determine, by the artificial intelligence large model, an answer quality of the interviewee to the interview question according to the answer data.

[0006] According to a third aspect, an electronic device is provided, including: at least one processor; and a memory communicatively connected with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described in any implementation manner of the first aspect.

[0007] According to a fourth aspect, a non-transitory computer-readable storage medium storing computer instructions is provided, and the computer instructions are used to enable a computer to perform the method described in any implementation manner of the first aspect.

[0008] According to a fifth aspect, a computer program product is provided, including: a computer program, which, when executed by a processor, implements the method described in any implementation manner of the first aspect.

[0009] According to the technology of the present disclosure, a large model-based simulated interview method and device are provided. The large model-based simulated interview method and device generate an interview question for an interviewee according to historical question and answer data of the interviewee in an interview process through an artificial intelligence large model. The interviewee's response data to the interview question is obtained. The response quality of the interviewee to the interview question is determined according to the response data through the artificial intelligence large model. Thus, the interview question is dynamically generated according to the historical question and answer data of the interviewee by using the artificial intelligence large model, the logicality of the interview question in the interview process is improved, the interactivity of the simulated interview process and the user is improved, and the simulated interview process is closer to a real interview scene. In addition, the response quality is determined according to the response data of the interview question in real time to provide real-time feedback to the interviewee, which helps to improve the interview skills of the interviewee and meets the simulated interview needs of the interviewee.

[0010] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0011] The accompanying drawings are used to better understand the present scheme and do not limit the present disclosure. Among them: Figure 1 is an exemplary system architecture diagram to which an embodiment according to the present disclosure can be applied; Figure 2 is a flowchart of one embodiment of the large model-based simulated interview method according to the present disclosure; Figure 3 is an interview logic flowchart according to the present embodiment; Figure 4 is a simulated interview process architecture timing diagram according to the present embodiment; Figure 5 is a schematic diagram of an application scenario of the large model-based simulated interview method according to the present embodiment; Figure 6 is a flowchart of another embodiment of the large model-based simulated interview method according to the present disclosure; Figure 7 is a structural diagram of one embodiment of the large model-based simulated interview device according to the present disclosure; Figure 8 is a structural schematic diagram of a computer system suitable for implementing the embodiments of the present disclosure. DETAILED DESCRIPTION

[0012] Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, which include various details of the embodiments of the present disclosure to assist in understanding them. These should be considered as merely exemplary. Thus, those skilled in the art will recognize that variations and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, in the interest of clarity, not all features of an actual implementation can be described in this description.

[0013] In the technical solutions of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved in the technical solutions comply with relevant laws and regulations and do not violate public order and good customs.

[0014] Figure 1 An exemplary architecture 100 of the large model-based simulated interview method and device to which the present disclosure can be applied is shown.

[0015] As shown in Figure 1 The system architecture 100 can include terminal devices 101, 102, 103, a network 104, and a server 105. The terminal devices 101, 102, 103 are communicatively connected to form a topological network, and the network 104 is used as a medium to provide communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 can include various connection types, such as wired, wireless communication links, or optical fiber cables, etc.

[0016] The terminal devices 101, 102, 103 can be hardware devices or software that support network connections for data interaction and data processing. When the terminal devices 101, 102, 103 are hardware, they can be various electronic devices that support network connections, information acquisition, interaction, display, processing, etc., including but not limited to smartphones, tablet computers, e-book readers, laptop computers, and desktop computers, etc. When the terminal devices 101, 102, 103 are software, they can be installed in the above-mentioned electronic devices. They can be implemented as multiple software or software modules for providing distributed services, or as a single software or software module. No specific limitation is made here.

[0017] The server 105 can be a server that provides various services, such as a background processing server that acquires a simulated interview request issued by a target object through a terminal device 101, 102, 103, performs a simulated interview on the interviewee through an artificial intelligence large model, and determines the response quality of the interviewee during the simulated interview. As an example, the server 105 can be a cloud server.

[0018] It should be noted that the server can be hardware or software. When the server is hardware, it can be implemented as a distributed server cluster composed of multiple servers, or as a single server. When the server is software, it can be implemented as multiple software or software modules (for example, software or software modules used to provide distributed services), or as a single software or software module. This is not specifically limited here.

[0019] It should also be noted that the large model-based simulated interview method provided by the embodiments of the present disclosure is generally executed by a server, but the possibility of being executed by a terminal device or by a server and a terminal device in cooperation with each other cannot be excluded. Accordingly, each part (for example, each unit) included in the large model-based simulated interview apparatus can be all arranged in the server, all arranged in the terminal device, or arranged in the server and the terminal device respectively.

[0020] It should be understood that Figure 1 The number of terminal devices, networks and servers in the system architecture is only illustrative. According to the needs of implementation, there can be any number of terminal devices, networks and servers. When the electronic device on which the large model-based simulated interview method runs does not need to perform data transmission with other electronic devices, the system architecture can only include the electronic device (for example, a terminal device or a server) on which the large model-based simulated interview method runs.

[0021] Please refer to Figure 2 , Figure 2 A flowchart of a large model-based simulated interview method provided by the embodiments of the present disclosure is shown in FIG. 2. In the flowchart 200, the following steps are included: In step 201, an artificial intelligence large model is used to generate an interview question for an interviewee according to historical question and answer data in an interview process of the interviewee.

[0022] In this embodiment, the execution subject (for example, the server in Figure 1 ) of the large model-based simulated interview method can use an artificial intelligence large model to generate an interview question for an interviewee according to historical question and answer data in an interview process of the interviewee.

[0023] An artificial intelligence large model (referred to as a large model) refers to a class of artificial intelligence models with a large number of parameters constructed by artificial neural networks, such as large language models, visual large models, multi-modal large models, and basic scientific large models, etc. Taking a large language model as an example, it is a large-scale language model constructed based on deep learning technology, mainly used for processing natural language processing tasks. It learns the patterns and structures of language by training on large-scale data, and can generate natural language text or understand natural language input. In this embodiment, a multi-modal large language model can be specifically used, which usually includes the following modules: Input module: receives user inputted multi-modal data such as text, voice, image, video, etc. such as historical question and answer data during the simulation of the interview process.

[0024] Preprocessing module: preprocesses the inputted multi-modal data, for example, preprocessing of text data represented by voice includes word segmentation, text cleaning, etc. operations, and converts the text into a form that can be processed by the model.

[0025] Encoding module: encodes the preprocessed multi-modal data into vector form so that the model can understand and process it. Common encoding methods include word embedding and encoder in Transformer architecture. Among them, word embedding includes Word2Vec (Words to Vector), GloVe (Global Vectors for Word Representation), etc.

[0026] Model module: the core part, usually based on deep learning architecture (such as Transformer), responsible for processing the encoded data vector, language understanding and generation. The model learns the complex patterns and semantic relationships of language through multi-level neural network structure.

[0027] Decoding module: decodes the output vector of the model into natural language text, image, video, and generates a reply to the user input. Decoding methods can include greedy decoding, beam search, etc.

[0028] Output module: outputs the decoded text in a form readable by the user, such as text, image, video displayed on the screen.

[0029] The interviewee sends a simulation interview request to the above execution subject through a terminal device, and the simulation interview request includes job requirement data and interviewee profile data (such as resume data). The artificial intelligence large model generates the first interview question in the interview process according to the job requirement data and the interviewee profile data. In the subsequent interview process, whenever a round of question and answer is completed, the artificial intelligence large model generates an interview question for the interviewee according to the historical question and answer data of the interviewee in the interview process.

[0030] The historical question and answer data can be all question and answer data up to the present, or question and answer data up to the present preset round, which can be flexibly set according to actual interview needs, and is not limited herein.

[0031] In order to improve the accuracy of the interview question and the adaptability of the interview question to the interviewee, the artificial intelligence large model can refer to the job requirement data and the interviewee profile data when generating the interview question in addition to the historical question and answer data.

[0032] As an example, first, for the job requirement data, the core elements are extracted and classified and labeled. For example, the requirements of "E-commerce operation position" are broken down into "hard skill class" (such as proficiency in using e-commerce platform background, data analysis ability), "soft skill class" (such as customer communication ability, activity planning execution force), "experience class" (such as more than 1 year of platform operation experience, with case of creating hit single product), and the priority of each element is marked (such as "data analysis ability" is the core requirement, and the priority is set to the highest). For the resume data, the key information extraction and association labeling are also carried out: on the one hand, the "ability matching points" (such as mentioning "familiar with using Excel for sales data review" in the resume, corresponding to "data analysis ability" in the job requirements), "experience association points" (such as "responsible for e-commerce big promotion activity execution" in the resume, corresponding to "activity planning execution force" in the job requirements); on the other hand, mark "to be verified information" (such as "participated in user growth project" in the resume, without specifying the specific role and results), "potential gap points" (such as "possessing short video marketing experience" in the job requirements, without mentioning related content in the resume).

[0033] After completing the data processing, the above structured job requirement elements and resume key information are input into the artificial intelligence big model, and the pre-set "interview question generation rule library" (including problem logic for different post types, such as "problem solving thought" type questions for technical positions, and "scene response" type questions for sales positions) is loaded to complete model initialization.

[0034] Then, based on the initialized structured data, the big model generates the first interview question around the "core requirements of the post and the key matching points in the resume", ensuring that the question focuses on core adaptability verification from the early stage of the interview. The specific generation logic is: first, filter the highest priority elements in the job requirements (such as "data analysis ability" of "e-commerce operation position"), then associate the "ability matching points" in the resume data corresponding to it (such as "familiar with using Excel for sales data review"), and then combine the problem templates in the "question generation rule library" for this type of post to generate the first question with verification and extension.

[0035] For example, if the core requirement of the post is data analysis ability, and the resume mentions "using Excel to review sales data", the big model will avoid generating simple "whether you will use Excel to analyze data" type closed questions, but generate open-ended questions: "In your resume, you mentioned using Excel to review sales data. Can you specifically describe an operation problem you discovered through data review, and how you adjusted the strategy based on the analysis?" In this way, the authenticity of "data analysis ability" in the resume is verified, and the interviewee is guided to expand the description, accumulating detailed data for subsequent question generation.

[0036] Then, whenever a round of questioning is completed (i.e. after the interviewee has finished answering the current interview question), the large model will first generate the next question based on the historical question and answer data, job requirement data, and profile data up to the end of this round of questioning, following the logic of "supplementary verification - deep mining - gap exploration", to ensure that the question chain is coherent and gradually deepens.

[0037] First, analyze the data of this round of questioning: extract the "key information points" (such as the answer to the first question above, mentioning "through analysis, it was found that the decline in average order value was due to the high proportion of low-price lead-in products, and the subsequent adjustment of the lead-in product and profit product ratio increased the average order value by 15%"), "information gaps" (such as not explaining "how to determine that the high proportion of lead-in products is the core reason"), and "new association points" (such as mentioning "synchronous optimization of product detail pages" in the answer, which can be associated with the "page optimization ability" in the job requirements), and merge these analysis results with the historical question and answer data (all previous rounds of question and answer analysis information) to form a real-time updated question and answer data pool.

[0038] Second, the large model generates the next question based on the question and answer data pool, job requirement data, and profile data.

[0039] If there is "to-be-verified information" in the question and answer data pool that has not been covered (such as the role of "participating in user growth projects" not being explained in the profile, and previous questions and answers not involving it), a "supplementary verification type" question is generated, such as: "You mentioned in your resume that you participated in the user growth project. Can you explain the specific work you were responsible for in that project, and which strategies you proposed had a practical impact on user growth?"

[0040] If the question and answer data pool has covered most of the "to-be-verified information", but there is "insufficient depth verification" of a certain core requirement of the job (such as the job requirement "event planning execution ability", and the previous answer only mentions "participating in e-commerce promotions" without explaining the difficulties encountered during the process), a "deep mining type" question is generated, such as: "When you were responsible for the execution of the e-commerce promotion event, what was the biggest execution difficulty you encountered? How did you coordinate resources to solve that difficulty and ensure that the event went online on time?"

[0041] If the question and answer data pool has verified most of the matching points, and there is a "potential gap point" (such as the job requirement "short video marketing experience", which is not mentioned in the profile or historical questions and answers), a "gap exploration type" question is generated to avoid directly questioning the gap, but instead uses the logic of "possibility mining" to guide the answer, such as: "Short video marketing in e-commerce operations is gradually playing a more prominent role in traffic acquisition. Have you ever been involved in related work? If you are required to take charge of the short video marketing board in the future, what aspects will you focus on to learn and practice?"

[0042] After generating the interview question, the interview question is fed back to the terminal device of the interviewee, and the terminal device is used to display the interview question to the interviewee.

[0043] With continued reference to Figure 3 , an interview logic flowchart based on a state machine-based workflow engine is shown.

[0044] To constrain the output of the large model to play the role of a professional interviewer rather than a casual chatbot, a structured interview strategy workflow engine is designed in this embodiment. The engine manages the entire interview process in the form of a state machine: The flow is divided into multiple atomic nodes, such as start_interview (opening and initialization), decide_next_step (decide next step), generate_question (generate question), ask_and_get_answer (ask and get answer), evaluate_answer (evaluate answer), end_interview (end interview and generate report), etc.

[0045] The transition between nodes is controlled by explicit conditions. For example, after the evaluate_answer node, the engine will determine whether the user's answer is a valid interview question and answer (is_valid_qa: True) or irrelevant chat / question (is_valid_qa: False). If the former, the flow returns to decide_next_step to continue the interview; if the latter, the flow may return to ask_and_get_answer to ask the user to answer again or clarify.

[0046] The entire workflow shares a context state, which contains user information, job information, historical question and answer records, and current interview round, etc. This allows the large model to fully utilize historical information when generating the next question or evaluation, ensuring the coherence and logic of the conversation.

[0047] In some optional implementations of the present embodiment, the above-mentioned execution subject can execute the above-mentioned step 201 in the following manner: First, in response to the interviewee's answer to the previous interview question, the interviewee's answer data is determined to be a valid answer, and it is determined whether the interviewee has completed the current interview stage. Among them, the interview stage flows according to a preset interview stage sequence.

[0048] In this implementation, for different positions or technical fields, a universally applicable interview stage sequence can be used, such as "self-introduction -> project digging -> technical digging -> open question and answer".

[0049] Different interview stages can be set according to the characteristics of different positions or interview fields. For example, for an embedded development position, the interview stage sequence is "underlying technology foundation verification -> core project detail digging -> technical difficulty disassembly and solution verification -> cross-team collaboration scenario review -> code logic and debugging idea investigation -> technical learning and future planning"; for a B (Business-to-Business, enterprise-to-enterprise) sales position, the interview stage sequence is "industry cognition icebreaker -> customer demand mining simulation -> solution matching elaboration -> objection handling drill -> business negotiation idea sorting -> professional motivation and stress resistance cognition".

[0050] For each interview stage in the interview stage sequence, the number of interview questions can be preset. If the number of interview questions for completing an interview stage are effectively answered by the interviewee, the interview stage is determined to be completed. For each interview stage in the interview stage sequence, the information points to be obtained can also be preset. If all the information points are obtained through the interviewee's answer data in one or more rounds, the interview stage is determined to be completed.

[0051] The artificial intelligence large model performs natural language understanding on the interview questions and the answer data to determine the degree of adaptation between them, so as to determine whether the answer data of an interview question is an effective answer. For example, for an interview question, the expected answer target point of the interviewee is determined, and when the answer data covers or includes most of the answer target points, it is identified as an effective answer.

[0052] Then, in response to completing the current interview stage, the next interview stage of the current interview stage in the interview stage sequence is generated according to the historical question and answer data.

[0053] In response to completing the current interview stage, the next interview stage of the current interview stage is determined from the interview stage sequence; the interview stage and the historical question and answer data are input into the artificial intelligence large model to generate an interview question adapted to the next interview stage.

[0054] As an example, the artificial intelligence large model can determine the stage target of the next interview stage (for example, the field skills and project experience of the interviewee expected to be understood), combine the stage target, the historical question and answer data, the interviewee's profile data and the position requirement data to determine the information expected to be answered by the interviewee, and then generate an interview question adapted to the next interview stage based on the information expected to be answered by the interviewee.

[0055] In the implementation, the simulation interview process based on the artificial self-capable large model is constrained by the sequence of interview stages, so that the simulation interview process ensures normal and orderly operation on the basis of dynamic flexibility of the artificial self-capable large model, which helps to improve the standardization and reliability of the simulation interview process.

[0056] In some optional implementations of the embodiment, the execution subject can execute the generation process of the interview question in the following manner: First, generate question generation prompt words for guiding the question generation process of the artificial intelligence large model according to the historical question and answer data and the next interview stage.

[0057] As an example, the artificial intelligence large model can determine the stage target of the next interview stage, combine the stage target, historical question and answer data, resume data of the interviewee, job requirement data, and explicit instructions (for example, combine the stage target, historical question and answer data, resume data of the interviewee, and job requirement data to generate an interview question suitable for the stage target), and obtain structured question generation prompt words. The question generation prompt words are used to guide the artificial intelligence large model to generate an interview question suitable for the next interview stage.

[0058] Then, generate an interview question suitable for the next interview stage according to the question generation prompt words.

[0059] Input the question generation prompt words into the artificial intelligence large model, and the artificial intelligence large model generates an interview question suitable for the next interview stage.

[0060] In the implementation, specific implementations of the artificial intelligence large model generating an interview question suitable for the next interview stage are provided, which helps to further improve the accuracy of the interview question and the degree of adaptation to the interview stage under the guidance of the prompt words.

[0061] In some optional implementations of the embodiment, the execution subject can execute the step 201 in the following manner: In response to the response quality of the interviewee to the previous interview question, the response data of the interviewee to the previous interview question is invalid answer, and an interview question representing a clarifying follow-up question for the previous interview question is generated according to the historical question and answer data.

[0062] In response to the response data of the interviewee to the previous interview question being invalid answer, it indicates that the execution subject has not obtained effective information based on the previous interview question, or the effective information obtained is very limited, and the execution subject needs to determine the target information to be clarified by the interviewee according to the response quality, and then generate an interview question representing a clarifying follow-up question for the previous interview question according to the target information to be clarified. In some optional implementations of the embodiment, the execution subject can execute the step 201 in the following manner:

[0063] For example, the last interview question is "Please describe in detail your process of handling the project", and the interviewee's answer data is irrelevant to the interview question, so the interview question representing the clarifying follow-up question for the last interview question needs to be generated.

[0064] In the invalid answer situation of the interviewee, the interview question representing the clarifying follow-up question for the last interview question needs to be further generated in the present implementation manner, so as to ask the interviewee, which helps to improve the completeness and abnormal situation handling capability of the simulated interview process.

[0065] In some optional implementation manners of the present embodiment, the above-mentioned execution subject can execute the above-mentioned interview question generation process in the following manner: First, in response to the fact that the interviewee's answer quality for the last interview question represents that the interviewee's answer data for the last interview question is an invalid answer, the problem generation prompt word for guiding the problem generation process of the artificial intelligence large model is generated according to the historical question and answer data and the current interview stage.

[0066] According to the historical question and answer data (especially the invalid answer data of the last interview question) and the current interview stage, the target information (the information expected to be answered by the interviewee based on the last interview question) not covered in the interviewee's answer data is determined, and the problem generation prompt word is generated in combination with the stage target of the interview stage and the target information not covered. The problem generation prompt word is used to guide the artificial intelligence large model to generate the interview question representing the clarifying follow-up question for the last interview question.

[0067] Then, according to the problem generation prompt word, the interview question representing the clarifying follow-up question for the last interview question is generated.

[0068] The problem generation prompt word is input into the artificial intelligence large model, and the artificial intelligence large model generates the interview question representing the clarifying follow-up question for the last interview question.

[0069] In the invalid answer situation of the interviewee, the prompt word for guiding the artificial intelligence large model to generate the interview question representing the clarifying follow-up question for the last interview question is generated in the present implementation manner, which further improves the accuracy of the interview question based on the prompt word.

[0070] In some optional implementation manners of the present embodiment, the above-mentioned execution subject can execute the above-mentioned step 201 in the following manner: in response to the fact that the current interview stage is not completed, the interview question representing the exploratory follow-up question for the last interview question is generated according to the historical question and answer data.

[0071] For example, if the number of interview questions output in the current interview stage does not reach the preset number of interview questions in the current interview stage, an interview question representing an exploratory follow-up question for the previous interview question can be generated according to the response data of the interviewee to the previous interview question, the job responsibility data, the profile data of the interviewee, and the stage target of the current interview stage, so as to continue to explore the relevant content in the response data and obtain more in-depth and comprehensive answers of the interviewee.

[0072] For example, if all the target information expected to be obtained in the current interview stage is not fully obtained, an interview question representing an exploratory follow-up question for the previous interview question can be generated according to the response data of the interviewee to the previous interview question, the job responsibility data, the profile data of the interviewee, and the unobtained target information, so as to continue to explore the relevant content in the response data and obtain more in-depth and comprehensive answers of the interviewee.

[0073] In the present implementation, for the case that the current interview stage is not completed, an interview question representing an exploratory follow-up question for the previous interview question is generated to continue the interview process of the current interview stage, which helps to improve the integrity and reliability of the simulated interview process.

[0074] In some optional implementations of the present embodiment, the above-mentioned execution subject can execute the above-mentioned interview question generation process in the following manner: First, in response to the current interview stage not being completed, a question generation prompt word for guiding the question generation process of the artificial intelligence large model is generated according to the historical question and answer data and the current interview stage.

[0075] In response to the current interview stage not being completed, the question generation prompt word can be obtained by combining the historical question and answer data, the uncompleted stage target in the current interview stage, and the explicit instruction (for example, generating an interview question representing an exploratory follow-up question for the previous interview question according to the historical question and answer data and the uncompleted stage target), which is used to guide the artificial intelligence large model to generate an interview question representing an exploratory follow-up question for the previous interview question.

[0076] Then, an interview question representing an exploratory follow-up question for the previous interview question is generated according to the question generation prompt word.

[0077] The question generation prompt word is input into the artificial intelligence large model, and the artificial intelligence large model generates an interview question representing an exploratory follow-up question for the previous interview question.

[0078] In the present implementation, for the case that the current interview stage is not completed, a prompt word is first generated according to the historical question and answer data and the current interview stage to guide the question generation process of the artificial intelligence large model, which helps to improve the accuracy of question generation.

[0079] At step 202, the response data of the interviewee to the interview question is acquired.

[0080] In this embodiment, the above execution subject can acquire the response data of the interviewee to the interview question.

[0081] With reference to the accompanying drawings still, Figure 4 , a schematic architecture timing diagram in the interview process is shown.

[0082] The user terminal application or Web in the terminal device of the interviewee can collect the voice data of the user and call the ASR (Automatic Speech Recognition) service to convert the voice data into text data; the terminal device transmits the text data to the backend service in the above execution subject.

[0083] In order to solve the data transmission problem between the terminal device and the above execution subject in a weak network environment, data communication can be carried out based on WebSocket long link.

[0084] Based on the WebSocket long link, real-time transmission and network disconnection retransmission can be realized between the client (corresponding to the terminal device) and the server (corresponding to the above execution subject). When the interviewee starts to speak, the client establishes a WebSocket long link with the server, and sends the collected audio stream data to the server in real time. The client has a network disconnection retransmission protocol built-in, which maintains a data packet queue to be confirmed. When the fragmented data packet is sent, the data packet or data packet identifier is added to the data queue. In response to receiving the acknowledgement message of the data packet from the server, the data packet or data packet identifier in the data packet queue is deleted. When the network is temporarily interrupted and then recovered, the client will automatically resend the data packet that has not been confirmed by the backend, ensuring the continuity and integrity of the audio data.

[0085] In order to ensure the reliability of data packet storage, the server stores the data packet in real time. When the server receives the data packet, it will be written into the object storage database in real time. A temporary file corresponding to the interview process of the interviewee is stored in the object storage database. This ensures that even if the interviewee exits abnormally in the middle of the interview, most of the recorded data has been persisted in the cloud, maximizing the protection of data security.

[0086] After a complete interview session, the server triggers an asynchronous task to transcode the original audio file (e.g., in PCM format) stored in object storage (e.g., to MP3 format) and securely save it as a complete interview record in the interviewee's network drive for easy review at any time.

[0087] In step 203, the AI large model determines the response quality of the interviewee to the interview question based on the response data.

[0088] In this embodiment, the above execution subject can perform data retrieval according to the adjusted query request to obtain the retrieval result.

[0089] As an example, the above execution subject can perform core dimension decomposition on the response data and quantitatively score each pair of dimensions to determine the response quality. Specifically, first, the AI large model performs information extraction and structural processing on the response data to first determine the "investigation core" of the current interview question. For example, if the question is "Please explain your specific process for handling customer complaints in the past", the model will first locate the investigation target of this question as "process integrity, problem solving logic, and customer communication awareness" three core dimensions, and extract the key information in the response data, including "whether to mention complaint reception, cause analysis, solution, and follow-up" process nodes, "whether to explain the specific operation of each link (such as "record complaint content through the work order system")", and "whether to mention "calm the customer's emotions" "avoid complaint escalation" and other communication-related expressions".

[0090] Then, the large model sets hierarchical evaluation standards for each investigation dimension and judges the compliance of the response data. For example, in the process integrity dimension, if the response contains "reception-analysis-solution-follow-up" full process, and each link has specific details, it is determined as "completely consistent"; if it lacks the "follow-up" link, but other links are complete, it is determined as "partially consistent"; if it only mentions "handling complaints" without explaining any process, it is determined as "inconsistent".

[0091] Next, the large model matches the corresponding score (e.g., "completely consistent" gets 8-10 points, "partially consistent" gets 4-7 points, and "inconsistent" gets 0-3 points) for each dimension evaluation result, and adjusts it with the aid of "language expression coherence". If the response has frequent stalls, repeated semantics (e.g., "then…then it is…is to answer the phone first"), or chaotic logic (e.g., reversed process order), deduct 1-2 points from the base score of each dimension; if the expression is smooth and the logic is clear, do not deduct points or even add 1 point (total score does not exceed the full score of the dimension).

[0092] Finally, the large model will integrate the scores of each dimension and the adjustment score of language expression to calculate the total score of response quality, generating the "total score + dimension feedback" of single question response quality, for example, "response quality total score 7.5 points (full score 10 points): process integrity 8 points, problem solving logic 7 points, customer communication awareness 8 points, and 0.5 points deducted for slight expression lag; Core conclusion: the response basically covers the key points, but the expression fluency needs to be improved."

[0093] As another example, the above execution subject can evaluate the response quality based on the combination of post demand matching and problem target verification mode. Specifically, first, the artificial intelligence large model will first associate the "post demand details" corresponding to the current interview question with the "problem target", such as if the interview post is "e-commerce operation post", the question is "please say a case of how you improve the sales of goods through data analysis", the model will first call the demand points of this post "need to have data analysis ability, sales optimization awareness, result-oriented thinking", and at the same time, it is clear that the target of this question is "to verify whether the candidate has a real data analysis case, and the case can reflect the actual driving effect on sales."

[0094] Then, the large model will perform bidirectional matching and verification on the response data and "post demand + problem target". First, it judges whether the response is "tight to the problem target" - that is, whether a "specific case" is provided (rather than a general "will use data tools"), whether the case contains "data source (such as'store back-end sales data'), analysis dimension (such as 'average order value, conversion rate, traffic source'), analysis conclusion (such as 'found that mobile end conversion rate is lower than personal computer end'), and 'action based on analysis (such as 'optimize mobile end product detail page')"; Then judge whether these contents match the post demand, such as whether it reflects the data analysis dimensions commonly used by e-commerce operation post (such as product click rate, add-to-cart rate), and whether the action is directed at sales increase (such as 'after optimizing the detail page, the mobile end conversion rate increased by 15%, driving sales growth by 8%').

[0095] Then, the large model will set three-level judgment standards for matching degree. If the answer case is real and specific, contains "data-analysis-action-result" whole chain, and each link is highly consistent with the data analysis demand of e-commerce operation post (such as using e-commerce exclusive data indicators, action consistent with operation routine strategy), it is determined as "high matching"; If the answer has a case, but lacks "data conclusion" or "sales result" (such as only saying "optimize the detail page after analyzing the data, but don't count the sales change"), or the case is weakly related to the e-commerce operation demand (such as using "offline store customer flow analysis" case to answer), it is determined as "medium matching"; If the answer has no specific case, only says "will improve sales through data analysis", or the case is irrelevant to the problem target (such as answering "how to do product inventory management"), it is determined as "low matching".

[0096] Next, the large model will supplement "detail feedback" for "high / medium / low matching" results. For high matching, it will point out the "highlight" in the answer (such as "can combine mobile conversion rate data to develop optimization strategies, meet the core needs of e-commerce operations"); for medium matching, it will clearly state the "missing points" (such as "the case lacks sales result verification, cannot completely prove the actual effect of data analysis"); for low matching, it will explain the "bias reasons" (such as "no specific case provided, cannot verify whether the data analysis ability meets the job requirements").

[0097] Finally, the large model will integrate the matching level and detail feedback to generate the answer quality conclusion, for example, "answer quality level: medium matching; core judgment: the answer provides a data analysis case, meets the problem target, but lacks sales result verification, slightly deviates from the'result-oriented' needs of e-commerce operations."

[0098] To further illustrate the simulation interview process, the following example is given: The interview initialization is completed, and the welcome word is output: Welcome, XX! I'm glad you came to today's interview. First of all, thank you for taking the time to communicate with us. This interview will be divided into four stages: first is self-introduction, then is project deep dive, followed by technical deep dive, and finally is open discussion. We will try to maintain a relaxed and natural atmosphere throughout the process, and hope you can relax and fully demonstrate your abilities and experience. Good luck with your interview!

[0099] The artificial intelligence large model makes decisions based on the following information and gets the decision results: Current state: stage=self_introduction (self-introduction), question number={'self_introduction': 0, 'project_deep_dive (project deep dive)': 0, 'technical_deep_dive (technical deep dive)': 0, 'open_discussion (open discussion)': 0}, history=0 The current stage'self_introduction' has completed question number: 0 Decision result: should_continu (whether to continue)=True, next_stage (next interview stage)=self_introduction, stage_switched (stage switch, indicating whether to enter the next interview stage)=False Routing decision details: - should_continue: True - is_interview_completed (whether the interview is completed): False - current_stage (current stage): self_introduction - stage_question_count (current question completion for each stage): {'self_introduction': 0, 'project_deep_dive': 0, 'technical_deep_dive': 0, 'open_discussion': 0} The decision result is GENERATE_QUESTION. Generate a self-introduction question: Please make a simple self-introduction, including your educational background, work experience, and main technical capabilities.

[0100] Interviewer: Please make a simple self-introduction, including your educational background, work experience, and main technical capabilities.

[0101] Interviewee: Hello, I graduated from XX School and am a new intern. I am here to interview for the backend development position. I am familiar with Golang and MySQL.

[0102] During the interviewee's response process, response data is collected. After the interviewee completes the response, the validity of the response data is evaluated. In response to validity, the question count for the current interview stage is updated: self_introduction 0 —> 1. At this time, the question count for all current interview stages is {'self_introduction': 1, 'project_deep_dive': 0, 'technical_deep_dive': 0, 'open_discussion': 0}.

[0103] Feedback: The interviewee's answer basically meets the requirements of the question, mentioning educational background (XX School), work experience (intern), and technical capabilities (Golang and MySQL). However, the content is too brief and lacks specific details. For example, there is no mention of the specific company of the internship, project experience, or technical depth. In addition, the answer is more colloquial and the structure is not clear enough.

[0104] Response quality: Relevance: 3 / 5 Completeness: 2 / 5 Technical accuracy: 3 / 5 Communication and expression: 3 / 5 The artificial intelligence large model continues to make decisions based on the following information, and obtains a decision result: Current state: stage=self_introduction, number of questions={'self_introduction': 1, 'project_deep_dive': 0, 'technical_deep_dive': 0, 'open_discussion': 0}, history=1 The current stage'self_introduction' has completed 1 questions. Decision result: should_continue=True, next_stage=project_deep_dive, stage_switched=True Stage switching: self_introduction —> project_deep_dive, reset new stage count.

[0105] By iterating the above process, the entire simulated interview process is finally completed.

[0106] In some optional implementations of the embodiment, the above execution subject can also perform the following operation: obtaining video data of the interviewee in the output process of the answer data.

[0107] The client not only needs to call the microphone to collect the answer data of the interviewee in the form of audio, but also needs to call the camera to synchronously collect the video data of the answer data output process. The video data includes the facial expressions and body movements of the interviewee.

[0108] The WebRTC (Real-Time Communication) protocol can be used for data transmission between the client and the server. The WebRTC protocol naturally supports synchronous transmission of audio and video, and has built-in mechanisms such as network jitter buffering, packet retransmission, and forward error correction, which can improve the security and reliability of data transmission in weak network environments.

[0109] In addition to real-time appending writing of the audio temporary file, the video stream slices also need to be written into the object storage database in parallel. After the interview is completed, an asynchronous task will combine the audio and video into a complete video file (such as an MP4 format) for the interviewee to review.

[0110] In the implementation, the above execution subject can execute the above step 203 by determining the answer quality based on the answer data and the video data through the artificial intelligence large model.

[0111] Taking a multi-modal large language model as an example, the response data and video data are input into the large language model, and the large language model outputs the response quality. For example, first, the audio data features (such as speech content, speech speed, pause frequency) and video data features (such as expression, body movement) of the interviewee's response are extracted, and the core of the current interview question (such as "project experience authenticity" and "logical expression ability") is determined.

[0112] Then, the content validity is judged by audio analysis. If the speech contains specific case details (such as "processing 300,000 user data with Python") and the speech speed is smooth without frequent stalls, it means that the content completeness and fluency meet the standards. On the contrary, if the content is empty (only "data analysis is done"), the pause is more than 3 seconds and there is no supplement, the content dimension will be deducted.

[0113] Next, the consistency is verified in combination with the video data. If the expression is natural (no avoiding eye contact and stiff smile), and the body has no defensive movements (such as crossing hands), it means that the response is highly reliable. If the expression is inconsistent with the content (saying "confidently complete the project" but frequently bowing), the reliability dimension will be deducted.

[0114] Finally, the content, fluency, and reliability scores are combined to generate a response quality conclusion (such as "content is complete and fluent, but the video shows slight nervousness, the reliability is medium, and the overall quality is qualified").

[0115] In the present implementation, the artificial intelligence large model determines the response quality in combination with the response data and video data of the interviewee, which helps to improve the accuracy and comprehensiveness of the response quality.

[0116] In some optional implementations of the present embodiment, the above-mentioned execution subject can perform the above-mentioned response quality determination process in the following manner: First step, according to the response data, determine the first sub-response quality representing the expression performance of the interviewee.

[0117] As an example, the response data is subjected to natural language understanding, and the expression performance of the interviewee is evaluated in multiple dimensions such as relevance, completeness, technical accuracy, and communication expression, to determine the first sub-response quality.

[0118] Second step, perform correlation analysis on the response data and video data, and determine the second sub-response quality representing the multi-modal behavior performance of the interviewee.

[0119] The video data is subjected to feature extraction to determine structured visual features, such as average eye contact rate (proportion of time the interviewee looks directly at the camera), dominant emotion (e.g., nervousness, naturalness), posture stability, gesture frequency, and oral fluency.

[0120] The correlation analysis of audio and video is the key to the value of video data, which can reveal the consistency and contradiction between the language content and non-verbal expression of the interviewee, and thus judge their true emotions and self-confidence. Specifically, the correlation analysis can be performed from the following multiple dimensions: Analysis dimension one: consistency analysis between the speech content represented by the response data and the emotion shown by the video data, used to judge whether the content said by the interviewee is consistent with the emotion conveyed by his facial expression and voice tone.

[0121] Specifically, the following multi-aspect emotion analysis is performed: Speech emotion analysis: by analyzing the prosodic features (pitch, energy, speech rate) of audio, the emotional attributes expressed are judged, such as positive, negative or neutral; Facial expression analysis through a visual large model, real-time recognition of user facial expressions (e.g. happy, surprised, focused, puzzled, nervous).

[0122] Text sentiment analysis: use a large language model (or traditional NLP model) to analyze the emotional color of the response text itself.

[0123] Based on the above multi-aspect situation analysis, the emotional consistency between multi-modal data is determined. For example, when the interviewee is talking about a "project that successfully overcomes great difficulties" (content positive), but the audio tone is flat and the facial expression is nervous, it can be inferred that the user may not be well prepared or lack confidence in this project. This provides an important basis for subsequent follow-up and final feedback.

[0124] Analysis dimension two: correlation between language fluency and visual behavior, used to identify the accompanying visual behavior of the interviewee when expressing unsmoothly (such as stuttering, repetition, using too many filler words), and locate the nervous point.

[0125] Specifically, the following multi-modal data analysis is performed: Audio analysis: identify filler words ("um", "ah", "that") and unnecessary pauses and repetitions in audio.

[0126] Visual analysis: synchronous detection of the visual behavior of the interviewee at this time, such as eye gaze shift (avoiding the camera), unconscious soothing actions such as touching the face / neck, and body shaking, etc.

[0127] If it is found that the interviewee has a "um..." habit every time he mentions the word "technical detail", accompanied by "looking up and to the right", it may be a strong signal that the interviewee lacks knowledge in this field.

[0128] Analysis dimension three: comprehensive evaluation of self-confidence and body language, used to comprehensively evaluate the overall self-confidence state exhibited by the interviewee when answering the question.

[0129] Specifically, the following multi-modal data analysis is performed: Audio features: stable speech rate, moderate volume, and falling intonation at the end of a sentence are usually associated with self-confidence.

[0130] Visual features include: Eye contact: consistently and steadily looking at the camera (representing the interviewer).

[0131] Body posture: through a posture estimation algorithm, determine whether the interviewee has an upright posture and a slight forward lean (indicating engagement).

[0132] Hand gestures: whether moderate and open gestures are used to assist in expression, rather than clenched fists or hidden hands.

[0133] The above execution subject can quantify these indicators, for example, by weighted summation between indicators, to calculate the self-confidence score of the question in real time, and present the trend changes in the final report.

[0134] The third step is to determine the answer quality by combining the first sub-answer quality and the second sub-answer quality.

[0135] The above execution subject can perform weighted summation on the first sub-answer quality and the second sub-answer quality to determine the answer quality.

[0136] In the present implementation, a determination method of answer quality in a multi-modal data situation is provided, and based on the correlation analysis of multi-modal data, the accuracy and analysis comprehensiveness of the answer quality are improved.

[0137] In some optional implementations of the present embodiment, the above execution subject can also perform the following operations: through an artificial intelligence large model, the interview quality of the entire interview process is determined in combination with the answer quality of the interviewee for each interview question in the interview process.

[0138] As an example, in combination with the answer quality of the interviewee for each interview question in the interview process, the correlation and logicality between the answer data are further analyzed to generate the interview quality.

[0139] Continuing to refer to Figure 5 , Figure 5is one schematic diagram 500 of an application scenario of the large model-based simulated interview method according to the embodiment. A target user 501 sends a simulated interview request to a server 503 through a terminal device 502. After the server receives the simulated interview request, the server first generates and displays a first interview question to the target user according to the job requirement data in the simulated interview request and the profile data of the interviewee, and the terminal device collects the response data of the target user to the first interview question. For the subsequent interview process, the server iteratively performs the following operations until the interview is completed: first, the server generates an interview question for the interviewee according to the historical question and answer data of the interviewee in the interview process through an artificial intelligence large model; then, the server obtains the response data of the interviewee to the interview question; finally, the server determines the response quality of the interviewee to the interview question according to the response data through the artificial intelligence large model.

[0140] In the embodiment, a large model-based simulated interview method is provided. The artificial intelligence large model is used to generate an interview question for an interviewee according to historical question and answer data of the interviewee in an interview process, obtain response data of the interviewee to the interview question, and determine response quality of the interviewee to the interview question according to the response data through the artificial intelligence large model. Thus, the artificial intelligence large model is used to dynamically generate an interview question according to historical question and answer data of an interviewee, improve the logicality of the interview question in an interview process, and improve the interactivity of a simulated interview process with a user, which is closer to a real interview scene. In addition, the response quality is determined according to the response data of the interview question, and feedback is provided to the interviewee in real time, which helps to improve the interview skills of the interviewee and meets the simulated interview needs of the interviewee.

[0141] In some optional implementation manners of the embodiment, the above execution subject can perform the following operations. In a first step, a dynamic performance parameter of the digital person is determined according to the interview question.

[0142] The dynamic performance parameter is, for example, a parameter for controlling the expression, mouth shape, and body movement of the digital person.

[0143] Analyze the type and atmosphere of the interview question, and determine the dynamic performance parameters of the digital person according to the type and atmosphere. For example, if the interview question is a basic cognitive question (such as "Please introduce your project experience"), the natural and friendly dynamic parameters need to be matched, the expression is set to a mild smile, the mouth corner is raised within the natural range, and excessive exaggeration is avoided; the mouth shape parameter is adjusted according to the pronunciation rhythm of the question text, to ensure that the opening and closing of the mouth shape is synchronized with the voice playback of each syllable, such as the transition from "xiang" to "mu" when pronouncing "project"; the body movement is set to lightly lift the right hand and naturally gesture, with small and slow movements to enhance the affinity. If it is a stress test type question (such as "What was the reason for your failure in leading a project"), the serious and steady parameters are matched, the expression is changed to neutral, the mouth corner is kept straight, and the eyes are focused; the mouth shape is kept clear but the rhythm is slightly slower, highlighting the seriousness of the question; the body movement is reduced, only keeping the upper body straight, avoiding unnecessary movements that distract the interviewer's attention.

[0144] In the second step, the digital person plays the interview question according to the dynamic performance parameters.

[0145] First, integrate the determined expression, mouth shape, and body movement parameters into unified execution instructions, and load the corresponding interview question voice materials. Before the digital person plays the question, perform a parameter pre-check to confirm that the expression animation and voice emotion are consistent (such as the smile of the affinity type question and the mild voice without conflict), the mouth shape frame is aligned with the voice timeline (such as the precise correspondence of each word's mouth shape change to the voice playback node), and the body movement start time is synchronized with the beginning of the question (such as starting to lightly lift the right hand after 1 second of basic question playback). After the verification is correct, start the digital person playback process, and monitor the parameter execution state in real time during the process. If a slight deviation is found between the mouth shape and the voice, immediately adjust the mouth shape parameter frame; if the body movement amplitude exceeds the set range, promptly correct the movement amplitude value, to ensure that the digital person plays the interview question smoothly according to the preset parameters, and returns to the initial neutral posture after the question is finished, waiting for the interviewee's response.

[0146] In the present implementation, the digital person plays the interview question, further improving the fit between the simulated interview process and the real interview process, and helping to further improve the interviewee's immersion in the interview process.

[0147] In some optional implementations of the present embodiment, the above-mentioned execution subject can also perform the following operations: According to the needs of the interviewee, determine the type of the digital person.

[0148] For example, if the interviewee wants to improve his or her stress resistance in the interview process, a digital person with a serious expression and harsh tone can be determined.

[0149] Alternatively, according to the selection operation of the interviewee, the type of the digital person is determined.

[0150] Based on the selection operation of the interviewee, the type of digital human can be determined from the list of digital humans, or based on the selection of attributes such as the appearance, gender, etc. of the digital human, the desired type of digital human is made.

[0151] In the present implementation, the execution subject can execute the first step by determining the dynamic performance parameters of the type of digital human according to the interview questions.

[0152] For example, after converting the interview question completion text into speech, the dynamic performance parameters such as expression, lip shape, body movement, etc. are determined in combination with the type of digital human.

[0153] In the present implementation, the type of digital human can be flexibly selected based on the needs or selection operations of the interviewee, which helps to further improve the adaptation of the interview process to the interviewee and the immersion of the interviewee in the simulated interview process.

[0154] In some optional implementations of the present embodiment, the execution subject can further perform the following operations: The first step is to generate a suggestion generation prompt word according to the response quality, wherein the suggestion generation prompt word is used to guide the artificial intelligence large model to generate a guidance suggestion for the output process of the response data; the second step is to generate a guidance suggestion for the output process of the response data by the artificial intelligence large model according to the suggestion generation prompt word.

[0155] As an example, first, the structured quantitative features of the multi-modal data corresponding to the response quality are determined. For example, the features corresponding to the audio data include: transcribed_text: the text content of the answer.

[0156] filler_word_count: the number of filler words.

[0157] speech_rate: average speech rate (words per minute).

[0158] pitch_variance: standard deviation of pitch variation.

[0159] The features corresponding to the video data include: eye_contact_percentage: the proportion of time spent looking at the lens area.

[0160] emotion_distribution: facial expression distribution map.

[0161] posture_assessment: posture assessment result.

[0162] gesture_classification: gesture classification statistics.

[0163] head_nod_count: number of nods as a signal of positive feedback.

[0164] In combination with the response data and the structured features described above, a suggestion generation prompt is obtained. The suggestion generation prompt is input into the artificial intelligence large model, and the artificial intelligence large model generates guidance suggestions for the output process of the response data.

[0165] To further improve the interactivity of the interview process, the digital person can feed back the guidance suggestions to the interviewee.

[0166] In the implementation mode, for each response data of the interviewee, the artificial intelligence large model gives targeted guidance suggestions, which helps to quickly improve the interview skills of the interviewee and further meets the simulation interview needs of the interviewee.

[0167] In some optional implementation modes of the embodiment, the above-mentioned execution subject can execute the above-mentioned first step to generate the suggestion generation prompt in the following manner: First, according to the response quality, the type of the point to be improved exhibited by the interviewee in the output process of the response data is determined by the artificial intelligence large model; then, according to the response quality and the type, the suggestion generation prompt is generated.

[0168] As an example, first review the core evaluation dimensions of the response quality (such as content integrity, logical fluency, credibility, and post matching degree), and extract the deduction items or deficiency descriptions of each dimension. If the response quality feedback is “content is empty and does not mention the specific operation details of the project”, it corresponds to the “content dimension-detail missing type” point to be improved; if the feedback is “frequent expression stalls, semantic logic is reversed”, it corresponds to the “expression dimension-logical confusion type”; if the feedback is “the answer does not combine the post requirements, and uses general words to deal with professional problems”, it corresponds to the “matching dimension-post adaptation deficiency type”; if the audio and video data is combined to find that “expression is nervous, body defense is contradictory to the confident expression of the answer”, it corresponds to the “presentation dimension-physical and mental coordination deficiency type”. At the same time, the model will exclude repeated or secondary deficiencies and focus on 1-2 core points to be improved to avoid too many types leading to subsequent suggestions being scattered.

[0169] The large model generates a suggestion generation prompt word according to the response quality level and the core point to be improved. The prompt word needs to clearly indicate the "target scene (for response improvement)", "core problem (point to be improved)", "suggestion direction (specific and operable optimization method)", and "post adaptation (adjust the suggestion focus according to the post requirements)". For example, if the response quality is "qualified (60 points)", the core point to be improved is "content dimension - detail missing type", and the post is an e-commerce operation post, the prompt word will be generated: "Please generate improvement suggestions for the response to the question 'Describe the activity marketing strategy' in the e-commerce operation post interview: 1. Need to supplement'specific marketing action details' (such as explaining 'adjusted direct channel keyword type' and 'promotion rules (full price / discount)'); 2. Need to add 'data evidence' (such as 'daily sales increased from 50 to 120, with a growth rate of 140% before and after the activity'); 3. Suggest combining the'result-oriented' characteristics of the e-commerce post to highlight the 'action-data' causal relationship, and avoid generalization."

[0170] If the response quality is "to be improved (45 points)", the core point to be improved is "expression dimension - lack of logical fluency", and the prompt word will emphasize "basic expression optimization": "Please generate improvement suggestions for the interview response: 1. Suggest using the 'STAR method' to organize the logic (first describe the'scene', then the 'task', 'action', and'result'); 2. Pre-set the response framework (such as silently reciting 'I will explain from the action and data aspects') to reduce delays; 3. Avoid semantic reversal, and organize the content in the order of'reasons before results' to adapt to the core requirement of 'clear information transmission' in the interview."

[0171] Finally, the large model will verify the guidance and pertinence of the prompt word to ensure that the suggestions are not general (such as not saying "add more details", but saying "add marketing action and data details"), strongly related to the point to be improved (such as suggestions for "lack of logical fluency" not involving "credibility optimization"), and matching the post characteristics (such as technical post suggestions focusing on "technical term accuracy" and sales post suggestions focusing on "communication appeal"), and finally outputting a prompt word that can be directly used to generate specific improvement suggestions.

[0172] In the present implementation, a specific generation method for generating a suggestion generation prompt word is provided, which improves the accuracy of the prompt word, and in turn helps to improve the accuracy and pertinence of the guidance suggestions.

[0173] With reference to Figure 6 , another embodiment of a large model-based simulated interview method according to the present disclosure is shown in a schematic flow 600. In flow 600, the following steps are included: Step 601, in response to the response quality of the interviewee to the previous interview question, the interviewee's response data to the previous interview question is determined to be a valid answer, and whether the interviewee has completed the current interview stage is determined.

[0174] wherein the interview stage flows according to the preset interview stage sequence.

[0175] Step 602, in response to completion of the current interview stage, generating an interview question suitable for the next interview stage in the interview stage sequence according to the historical question and answer data.

[0176] Step 603, in response to the response quality of the interviewee to the previous interview question indicating that the response data of the interviewee to the previous interview question is invalid answer, generating an interview question representing a clarifying follow-up question for the previous interview question according to the historical question and answer data.

[0177] Step 604, in response to the current interview stage not being completed, generating an interview question representing an exploratory follow-up question for the previous interview question according to the historical question and answer data.

[0178] Step 605, determining the dynamic performance parameter of the digital human according to the interview question.

[0179] Step 606, controlling the digital human to play the interview question according to the dynamic performance parameter.

[0180] Step 607, obtaining the response data of the interviewee to the interview question and obtaining the video data of the interviewee in the output process of the response data.

[0181] Step 608, determining the response quality according to the response data and the video data by the artificial intelligence large model. Step 609, generating a suggestion generation prompt word according to the response quality.

[0182] wherein the suggestion generation prompt word is used to guide the artificial intelligence large model to generate a guidance suggestion for the output process of the response data.

[0183] Step 610, generating a guidance suggestion for the output process of the response data according to the suggestion generation prompt word by the artificial intelligence large model.

[0184] In the simulated interview process of the interviewee, the above-mentioned execution subject can iteratively execute the above-mentioned steps until the simulated interview process is completed.

[0185] Compared with the flow 200, the flow 600 of the large model-based simulated interview method in this embodiment specifically illustrates the generation process of the interview question, the question playing process based on the digital person, the generation process based on the guidance suggestion, improves the logic of the interview question in the interview process, and the interactivity of the simulated interview process with the user, and is closer to the real interview scene; and the response quality is determined in real time according to the response data of the interview question, the guidance suggestion is generated to feedback to the interviewee in real time, which helps to improve the interview skills of the interviewee, and meets the simulated interview needs of the interviewee.

[0186] With reference to the above-mentioned method shown in the figures, Figure 7 as an implementation of the above-mentioned method, the present disclosure provides one embodiment of a large model-based simulated interview device, which corresponds to the method embodiment shown in Figure 2 , and the system can be specifically applied to various electronic devices.

[0187] As shown in Figure 7 , the large model-based simulated interview device 700 comprises: a question generation unit 701 configured to generate an interview question for an interviewee by an artificial intelligence large model according to historical question and answer data in an interview process of the interviewee; an answer acquisition unit 702 configured to acquire response data of the interviewee to the interview question; and a quality evaluation unit 703 configured to determine the response quality of the interviewee to the interview question according to the response data by the artificial intelligence large model.

[0188] In some optional implementation manners of the present embodiment, the question generation unit 701 is further configured to: in response to the response quality of the interviewee to the previous interview question indicating that the response data of the interviewee to the previous interview question is a valid answer, determine whether the interviewee completes the current interview stage, wherein the interview stage flows according to a preset interview stage sequence; and in response to completing the current interview stage, generate an interview question for a next interview stage in the interview stage sequence according to the historical question and answer data.

[0189] In some optional implementation manners of the present embodiment, the question generation unit 701 is further configured to: generate a question generation prompt word for guiding the question generation process of the artificial intelligence large model according to the historical question and answer data and the next interview stage; and generate an interview question adapted to the next interview stage according to the question generation prompt word.

[0190] In some optional implementation manners of the present embodiment, the question generation unit 701 is further configured to: in response to the response quality of the interviewee to the previous interview question indicating that the response data of the interviewee to the previous interview question is an invalid answer, generate an interview question indicating a clarifying follow-up question for the previous interview question according to the historical question and answer data.

[0191] In some optional implementations of the embodiment, the question generation unit 701 is further configured to: in response to the response quality of the interviewee to the previous interview question representing that the response data of the interviewee to the previous interview question is an invalid answer, generate, according to the historical question and answer data and the current interview stage, a question generation prompt word for guiding the question generation process of the artificial intelligence large model; and generate, according to the question generation prompt word, an interview question representing a clarifying follow-up question to the previous interview question.

[0192] In some optional implementations of the embodiment, the question generation unit 701 is further configured to: in response to the current interview stage being incomplete, generate, according to the historical question and answer data, an interview question representing an exploratory follow-up question to the previous interview question.

[0193] In some optional implementations of the embodiment, the question generation unit 701 is further configured to: in response to the current interview stage being incomplete, generate, according to the historical question and answer data and the current interview stage, a question generation prompt word for guiding the question generation process of the artificial intelligence large model; and generate, according to the question generation prompt word, an interview question representing an exploratory follow-up question to the previous interview question.

[0194] In some optional implementations of the embodiment, the above-described apparatus further includes a video acquisition unit (not shown in the figure) configured to acquire video data of the interviewee in the output process of the response data; and the quality evaluation unit 703 is further configured to: determine, by the artificial intelligence large model, the response quality according to the response data and the video data.

[0195] In some optional implementations of the embodiment, the quality evaluation unit 703 is further configured to: determine, according to the response data, a first sub-response quality representing the expression performance of the interviewee; perform correlation analysis on the response data and the video data to determine a second sub-response quality representing the multi-modal behavior performance of the interviewee; and combine the first sub-response quality and the second sub-response quality to determine the response quality.

[0196] In some optional implementations of the embodiment, the above-described apparatus further includes a question playing unit (not shown in the figure) configured to: determine, according to the interview question, a dynamic performance parameter of the digital person; and control the digital person to play the interview question according to the dynamic performance parameter.

[0197] In some optional implementations of the embodiment, the device further includes a digital person selection unit (not shown in the figure) configured to determine the type of the digital person according to the needs of the interviewee, or determine the type of the digital person according to the selection operation of the interviewee, and the question playing unit is further configured to determine the dynamic performance parameters of the type of the digital person according to the interview questions.

[0198] In some optional implementations of the embodiment, the device further includes a guidance unit (not shown in the figure) configured to generate a suggestion generation prompt word according to the answer quality, wherein the suggestion generation prompt word is used to guide the artificial intelligence large model to generate a guidance suggestion for the output process of the answer data, and the guidance unit is further configured to generate the guidance suggestion for the output process of the answer data according to the suggestion generation prompt word through the artificial intelligence large model.

[0199] In some optional implementations of the embodiment, the guidance unit is further configured to determine the type of the point to be improved exhibited by the interviewee in the output process of the answer data according to the answer quality through the artificial intelligence large model, and generate the suggestion generation prompt word according to the answer quality and the type.

[0200] In the embodiment, a large model-based simulated interview device is provided. The artificial intelligence large model is used to generate an interview question for an interviewee according to historical question and answer data of the interviewee in an interview process, obtain answer data of the interviewee to the interview question, and determine an answer quality of the interviewee to the interview question according to the answer data through the artificial intelligence large model. Thus, the artificial intelligence large model is used to dynamically generate an interview question according to historical question and answer data of an interviewee, improve the logicality of the interview question in the interview process, and improve the interactivity of the simulated interview process with the user, which is closer to a real interview scene. In addition, the answer quality is determined in real time according to the answer data of the interview question, so as to provide real-time feedback to the interviewee, which helps to improve the interview skills of the interviewee and meet the simulated interview needs of the interviewee.

[0201] According to the embodiments of the present disclosure, the present disclosure further provides an electronic device, which includes at least one processor and a memory connected with the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to implement the large model-based simulated interview method described in any of the above embodiments.

[0202] According to the embodiments of the present disclosure, the present disclosure further provides a readable storage medium storing computer instructions for enabling a computer to implement the large model-based simulated interview method described in any of the above embodiments when the computer executes the computer instructions.

[0203] The embodiment of the present disclosure provides a computer program product, which can realize the large model-based simulated interview method described in any of the above embodiments when executed by a processor.

[0204] Figure 8 A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, tablets, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the present disclosure described and / or claimed in this document.

[0205] As Figure 8 shown, the device 800 includes a computing unit 801 that can perform various appropriate actions and processes in accordance with a computer program stored in a read-only memory (ROM) 802 or a computer program loaded into a random access memory (RAM) 803 from a storage unit 808. Various programs and data required for the operation of the device 800 can also be stored in the RAM 803. The computing unit 801, the ROM 802, and the RAM 803 are connected to each other through a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0206] Various components in the device 800 are connected to the I / O interface 805, including: an input unit 806, such as a keyboard, a mouse, etc.; an output unit 807, such as various types of displays, speakers, etc.; a storage unit 808, such as a magnetic disk, an optical disk, etc.; and a communication unit 809, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 809 allows the device 800 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunication networks.

[0207] The computing unit 801 can be various general and / or special purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs various methods and processes described above, such as the large model-based simulated interview method. For example, in some embodiments, the large model-based simulated interview method can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device 800 via the ROM 802 and / or the communication unit 809. When the computer program is loaded onto the RAM 803 and executed by the computing unit 801, one or more steps of the large model-based simulated interview method described above can be performed. Alternatively, in other embodiments, the computing unit 801 can be configured to perform the large model-based simulated interview method by any other suitable means, such as by means of firmware.

[0208] Various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0209] Program code for carrying out methods of the present disclosure can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, implements the functions / acts specified in the flowcharts and / or block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine and partially on a remote machine or entirely on a remote machine or server.

[0210] In the context of this disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium will include one or more lines of electrical connections, portable computer disks, hard disk drives, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or Flash memory), optical fibers, portable compact disc read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0211] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0212] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0213] The computer system can include clients and servers. This relationship can be. The servers are generally remote from the clients with the computers. The interaction can be carried out over the communication network. The relationship can be a client-server relationship over the communications network. The servers can be cloud servers, also known as cloud computing servers or cloud hosts, which are a host product in the cloud computing service system to solve the defects of large management difficulty and weak business scalability in traditional physical hosts and virtual private server (VPS, Virtual Private Server) services. They can also be servers of distributed systems or servers combined with blockchains.

[0214] According to the technical scheme of the embodiment of the present disclosure, a simulation interview method and device based on a large model are provided. The artificial intelligence large model generates an interview question for an interviewee according to historical question and answer data of the interviewee in an interview process. The response data of the interviewee to the interview question is obtained. The artificial intelligence large model determines the response quality of the interviewee to the interview question according to the response data. Thus, the artificial intelligence large model dynamically generates an interview question according to the historical question and answer data of the interviewee, improves the logicality of the interview question in the interview process, and improves the interactivity of the simulation interview process and the user, which is closer to the real interview scene. In addition, the response quality is determined according to the response data of the interview question in real time to provide real-time feedback to the interviewee, which helps to improve the interview skills of the interviewee and meets the simulation interview needs of the interviewee.

[0215] It should be understood that the various forms of the flow shown above can be used to reorder, add or delete steps. For example, the steps described in the present disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical scheme provided by the present disclosure can be achieved, which is not limited herein.

[0216] The above detailed description does not constitute a limitation on the protection scope of the present disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements and improvements within the spirit and principles of the present disclosure should be included in the protection scope of the present disclosure.

Claims

1. A large model-based simulated interview method, comprising: generating, by an artificial intelligence large model, an interview question for an interviewee according to historical question-and-answer data in an interview process of the interviewee; obtaining answer data of the interviewee to the interview question; determining, by the artificial intelligence large model, an answer quality of the interviewee to the interview question according to the answer data.

2. The method of claim 1, wherein, The generating, by an artificial intelligence large model, an interview question for an interviewee according to historical question-and-answer data in an interview process of the interviewee, comprises: determining whether the interviewee completes a current interview stage in response to the answer quality of the interviewee to a previous interview question indicating that the answer data of the interviewee to the previous interview question is a valid answer, wherein the interview stages flow according to a preset interview stage sequence; generating, in response to completing the current interview stage, an interview question adapted to a next interview stage of the current interview stage in the interview stage sequence according to the historical question-and-answer data.

3. The method of claim 2, wherein, The generating, in response to completing the current interview stage, an interview question adapted to a next interview stage of the current interview stage in the interview stage sequence according to the historical question-and-answer data, comprises: generating, according to the historical question-and-answer data and the next interview stage, a question generation prompt word for guiding a question generation process of the artificial intelligence large model; generating, according to the question generation prompt word, an interview question adapted to the next interview stage.

4. The method of claim 2, wherein, The generating, by an artificial intelligence large model, an interview question for an interviewee according to historical question-and-answer data in an interview process of the interviewee, further comprises: generating, in response to the answer quality of the interviewee to the previous interview question indicating that the answer data of the interviewee to the previous interview question is an invalid answer, an interview question representing a clarifying follow-up question for the previous interview question according to the historical question-and-answer data.

5. The method of claim 4, wherein, The generating, in response to the answer quality of the interviewee to the previous interview question indicating that the answer data of the interviewee to the previous interview question is an invalid answer, an interview question representing a clarifying follow-up question for the previous interview question according to the historical question-and-answer data, comprises: generating, in response to the answer quality of the interviewee to the previous interview question indicating that the answer data of the interviewee to the previous interview question is an invalid answer, a question generation prompt word for guiding a question generation process of the artificial intelligence large model according to the historical question-and-answer data and the current interview stage; generating, according to the question generation prompt word, an interview question representing a clarifying follow-up question for the previous interview question.

6. The method of claim 2, wherein, The generating, by an artificial intelligence large model, an interview question for an interviewee according to historical question-and-answer data in an interview process of the interviewee, further comprises: generating, in response to not completing the current interview stage, an interview question representing an exploratory follow-up question for the previous interview question according to the historical question-and-answer data.

7. The method of claim 6, wherein, The generating, in response to not completing the current interview stage, an interview question representing an exploratory follow-up question for the previous interview question according to the historical question-and-answer data, comprises: in response to the current interview stage being incomplete, generating, according to the historical question and answer data and the current interview stage, a question generation prompt word for guiding a question generation process of the artificial intelligence large model; generating, according to the question generation prompt word, an interview question representing an exploratory follow-up question for the previous interview question.

8. The method of any one of claims 1-7, wherein, Further comprising: obtaining video data of the interviewee in an output process of the answer data; and the determining, by the artificial intelligence large model, the answer quality of the interviewee to the interview question according to the answer data comprises: determining, by the artificial intelligence large model, the answer quality according to the answer data and the video data.

9. The method of claim 8, wherein, the determining, according to the answer data and the video data, the answer quality comprises: determining, according to the answer data, a first sub-answer quality representing an expression performance of the interviewee; performing a correlation analysis on the answer data and the video data to determine a second sub-answer quality representing a multi-modal behavior performance of the interviewee; combining the first sub-answer quality and the second sub-answer quality to determine the answer quality.

10. The method of any one of claims 1-9, wherein, Further comprising: determining a dynamic performance parameter of a digital person according to the interview question; controlling the digital person to play the interview question according to the dynamic performance parameter.

11. The method of claim 10, wherein, Further comprising: determining a type of the digital person according to a requirement of the interviewee; or determining a type of the digital person according to a selection operation of the interviewee; and the determining, according to the interview question, a dynamic performance parameter of a digital person comprises: determining a dynamic performance parameter of the type of digital person according to the interview question.

12. The method of any one of claims 1-11, wherein, Further comprising: generating a suggestion generation prompt word according to the answer quality, wherein the suggestion generation prompt word is used to guide the artificial intelligence large model to generate a guidance suggestion for an output process of the answer data; generating, by the artificial intelligence large model, a guidance suggestion for an output process of the answer data according to the suggestion generation prompt word.

13. The method of claim 12, wherein, the generating, according to the answer quality, a suggestion generation prompt word comprises: determining, by the artificial intelligence large model, a type of a point to be improved exhibited by the interviewee in the output process of the answer data according to the answer quality; generating the suggestion generation prompt word according to the answer quality and the type.

14. A large model-based simulated interview device, comprising: a question generation unit configured to generate, by an artificial intelligence large model, an interview question for an interviewee according to historical question and answer data in an interview process of the interviewee; an answer acquisition unit configured to acquire answer data of the interviewee to the interview question; a quality evaluation unit configured to determine, by the artificial intelligence large model, an answer quality of the interviewee to the interview question according to the answer data.

15. An electronic device, comprising: comprise: at least one processor; and a memory connected in communication with the at least one processor; wherein, The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-14.

16. A non-transitory computer-readable storage medium having stored thereon computer instructions, wherein, The computer instructions are for causing the computer to perform the method of any one of claims 1-14.

17. A computer program product, comprising: A computer program which, when executed by a processor, implements the method of any one of claims 1-14.

Images