Digital person business card interaction method, electronic equipment and storage medium

Abstract

The invention provides a digital person business card interaction method, electronic equipment and a storage medium, and belongs to the technical field of data processing, and the method comprises the steps: responding to a business card viewing instruction, and outputting a target digital person business card associated with the business card viewing instruction; receiving an inquiry instruction based on the target digital person business card, and judging whether the inquiry instruction has a corresponding standard problem or not; if yes, inputting an information set associated with the standard question and the target digital person business card into a pre-trained information desensitization large model; if not, calling a preset search engine to search the inquiry instruction, determining a search data set based on a search result, and inputting the inquiry instruction, the target digital person business card and the search data set into a pre-trained information desensitization large model; and obtaining a desensitization answer output by the information desensitization large model, and outputting the desensitization answer. According to the method, the interaction efficiency of the recruitment process can be remarkably improved, the hysteresis of information transmission is reduced, and the personal information security is protected.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to an interactive method, electronic device and storage medium for a digital human business card. Background Art

[0002] In the current recruitment process, if the employer wants to get to know the job seeker in depth, the two parties can usually only interact by sending messages. However, this method is not only not timely enough, but may also lead to a lag in information transmission. Although telephone communication can improve communication efficiency, for job seekers, repeating the same content when facing multiple employers will consume a lot of time and energy.

[0003] Therefore, how to improve the interactive efficiency of the recruitment process is a technical problem that needs to be solved urgently. Summary of the invention

[0004] The purpose of this application is to provide an interactive method, electronic device and storage medium for digital human business cards to solve the above problems.

[0005] To achieve the above objectives, in a first aspect, the present application proposes a digital business card interaction method, the method comprising: In response to a business card viewing instruction, outputting a target digital person business card associated with the business card viewing instruction, wherein the target digital person business card includes an image display area and a text introduction area; Receiving an inquiry instruction based on the target digital human business card, and determining whether the inquiry instruction has a corresponding standard question; If there is a standard question corresponding to the inquiry instruction, input the standard question and the information set associated with the target digital person business card into the pre-trained information desensitization model to obtain a desensitized answer output by the information desensitization model; If there is no standard question corresponding to the query instruction, calling a preset search engine to search the query instruction, and determining a search data set based on the search results, inputting the query instruction, the target digital person business card and the search data set into a pre-trained information desensitization model to obtain a desensitized answer output by the information desensitization model; Obtain the desensitized answer output by the information desensitization model, and output the desensitized answer.

[0006] In some implementations, determining whether the query instruction has a corresponding standard question includes: Acquire an information set associated with the target digital person's business card, the information set including a basic information set and a question-and-answer information set; Calculating the semantic similarity between the query instruction and each question in the question set; The calculated semantic similarity is compared with the preset similarity to determine whether the query instruction has a corresponding standard question, wherein the question with a semantic similarity greater than the preset similarity is the standard question corresponding to the query instruction.

[0007] In some implementations, before outputting the target digital person business card associated with the business card viewing instruction in response to the business card viewing instruction, the method includes: Receive a target digital person business card generation instruction, the target digital person business card generation instruction includes the target digital person business card style type and resume information, the target digital person business card style type includes the target digital person style type and business card layout style type; Inputting the resume information into a pre-trained information extraction model, wherein the pre-trained information extraction model extracts and outputs the portrait information and standard information in the resume information; Based on the style type of the target digital human and the portrait information, generate a target digital human image; Based on the target digital human image, the standard information and the business card typesetting style type, a target digital human business card is generated and output.

[0008] In some embodiments, before inputting the information set associated with the standard question and the target digital person's business card into the pre-trained information desensitization model, or inputting the query instruction, the target digital person's business card and the search data set into the pre-trained information desensitization model, the process includes: Acquire an initial training data set, wherein the initial training data set includes initial training data of job seekers of various industries, various positions, and various experience levels, and the initial training data includes resume information data and conversation data; Inputting the resume information, the conversation data and the instruction data into the information desensitization big model to train the information desensitization big model to output answer data to the instruction data based on the resume information and the conversation data; Obtaining a standard answer corresponding to the instruction data, and calculating an answer loss based on the answer data and the standard answer; Based on the answer loss, the information desensitization model is feedback updated by using elastic weight sharing until the training is completed.

[0009] In some embodiments, the inputting the resume information, the conversation data, and the instruction data into the information desensitization big model to train the information desensitization big model to output a model answer to the instruction data based on the resume information and the conversation data includes: Marking sensitive information in the resume information and the conversation data; The resume information and conversation data marked with sensitive information are input into the information desensitizing model to train the information desensitizing model to output a model response to the instruction data based on the resume information and the conversation data, wherein the sensitive information in the model response will be replaced with desensitized information.

[0010] In some implementations, the information desensitization model is feedback updated based on the answer loss by elastic weight sharing until the training is completed, including: Constructing a reinforcement learning training data set, wherein the reinforcement learning training data set includes the instruction data, a model answer corresponding to the instruction data, a standard answer corresponding to the instruction data, and an answer evaluation index; Based on the reinforcement learning training data set, a reinforcement learning environment is set, and a state space, an action space and a reward function are defined in the reinforcement learning environment, wherein the state space includes instruction data and a model answer corresponding to the instruction data, the action space includes an answer adjustment strategy, and the reward function is used to calculate a reward signal based on a standard answer corresponding to the instruction data and the answer evaluation index; In the reinforcement learning environment, the information desensitization model is updated based on the reward signal until the reinforcement learning training is completed.

[0011] In some embodiments, in the reinforcement learning environment, updating the information desensitization model based on the reward signal includes: Based on the specified instruction data in the state space and the model answer corresponding to the specified instruction data, selecting an answer adjustment strategy in the action space; Applying the selected answer adjustment strategy to the trained information desensitization model to generate a new model answer corresponding to the specified instruction data; Calculating a reward signal based on the new model answer corresponding to the specified instruction data, the standard answer corresponding to the specified instruction data and the reward function; Based on the reward signal, the information desensitization model is updated.

[0012] In some embodiments, constructing a reinforcement learning training dataset comprises: Constructing a reinforcement learning training data set, and dividing the reinforcement learning training data set into multiple subsets; The step of setting a reinforcement learning environment based on the reinforcement learning training data set includes: Create a corresponding reinforcement learning environment for each subset, where each reinforcement learning environment corresponds to a training process; In the reinforcement learning environment, updating the information desensitization model based on the reward signal includes: In each reinforcement learning environment, calculate the gradient information of each training process; The information desensitization model is updated by aggregating the gradient information of all training processes.

[0013] In a second aspect, the present application further proposes an electronic device, comprising: one or more processors; The memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors are enabled to execute the digital personal business card interaction method as described above.

[0014] In a third aspect, the present application further proposes a storage medium, wherein the storage medium stores executable instructions, and when the instructions are executed by a processor, the processor executes the interactive method of the digital human business card as described above.

[0015] Compared with the prior art, the beneficial effects of this application include: On the one hand, the digital business card of this application can be integrated into a variety of recruitment platforms and social media to achieve multi-channel synchronization of information and improve the efficiency of job information dissemination. On the other hand, the digital business card of this application can provide a more personalized display method, which is convenient for the recruiter to have a more intuitive understanding of the job seeker, and improves the efficiency and accuracy of the transmission of the job seeker's personal information. On the third hand, the digital business card of this application can respond to the recruiter's inquiry instructions immediately without waiting for the job seeker's reply, reducing the time cost of communication. On the fourth hand, for preset standard questions, this application can directly call the pre-trained information desensitization model to quickly generate answers, avoid the situation of repeatedly answering the same questions for job seekers, and save the time and energy of job seekers. On the fifth hand, for non-standard questions, this application can obtain relevant information through search engines, and generate answers in combination with pre-trained large models, ensuring the flexibility and wide applicability of the application of this application. On the sixth hand, the answers output by this application are desensitized answers, which can provide accurate and useful information to the recruiter under the premise of protecting the privacy of job seekers, further improving the interactive efficiency of the recruitment process. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope of the present application.

[0017] Figure 1 A flowchart of an embodiment of the interactive method of the digital human business card of the present application; Figure 2 A partial flow chart of an embodiment of the interactive method of the digital human business card of the present application; Figure 3 A flowchart of an embodiment of the interactive method of the digital human business card of the present application; Figure 4 A flowchart of an embodiment of the interactive method of the digital human business card of the present application; Figure 5 A flowchart of an embodiment of the interactive method of the digital human business card of the present application; Figure 6 A flowchart of an embodiment of the interactive method of the digital human business card of the present application; Figure 7 It is a schematic diagram of the structure of an electronic device involved in the interactive method of a digital human business card in any embodiment of the present application. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0019] All terms (including technical and scientific terms) used in this application have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used here should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0020] For example, the terms "first", "second", etc. used in this application may be used in this article to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish a first element from another element. For example, without departing from the scope of this application, a first state management container may be referred to as a second state management container, and similarly, a second state management container may be referred to as a first state management container. Both the first state management container and the second state management container are state management containers, but they are not the same state management container.

[0021] For another example, the terms “include”, “comprising”, etc. used in this application indicate the existence of features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0022] The recruitment-related data involved in this application has been obtained with full consent and authorization, and the collection, use and processing of relevant information complies with the relevant laws, regulations and standards of relevant countries and regions.

[0023] As mentioned above, in the current recruitment process, if the recruiter wants to get to know the job seeker in depth, the two parties can usually only interact by sending messages. However, this method is not only not timely enough, but may also cause a lag in information transmission. Although telephone communication can improve communication efficiency, for job seekers, repeating the same content when facing multiple recruiters will take up a lot of time and energy. Therefore, how to improve the interactive efficiency of the recruitment process is a technical problem that needs to be solved urgently. To this end, the present application proposes an interactive method, electronic device and storage medium for digital business cards, which can not only significantly improve the interactive efficiency of the recruitment process, reduce the lag in information transmission, and protect personal information security, but also improve user experience, reduce recruitment costs, and support large-scale applications.

[0024] like Figure 1 As shown, the interactive method of the digital human business card of this embodiment includes the following steps: Step S10, in response to the business card viewing instruction, outputting the target digital human business card associated with the business card viewing instruction, wherein the target digital human business card includes an image display area and a text introduction area.

[0025] The business card viewing instruction in this embodiment refers to a command for the user to trigger the system to display a specific digital person business card in some way. These instructions can come from a variety of terminals and application scenarios, including but not limited to smart phones, personal computers, robots, and other electronic devices with data processing, network communication, program running functions, and display screens. The specific online implementation form can be a website, mobile application (APP), WeChat applet, H5 page, API interface, etc. For example, the recruiter clicks on the avatar or name of a job seeker on the APP, applet or web page of the recruitment platform. The electronic device receives the business card viewing instruction, retrieves the digital person business card information of the job seeker corresponding to the business card viewing instruction from the database, and outputs the target digital person business card of the job seeker on the platform. For another example, the job seeker sends the link of the digital person business card to the recruiter by email or shares it on social media or other communication platforms. The recruiter clicks on the link and is redirected to a web page to display the target digital person business card.

[0026] In this embodiment, the target digital human business card has an image display area and a text introduction area. The image display area can be used to display the job seeker's photo, digital human image, real-life video or digital human video. The text introduction area can contain detailed information such as the job seeker's name, position, work experience, educational background, skill certificates, contact information, etc.

[0027] Step S20: receiving an inquiry instruction based on the target digital human business card.

[0028] In this embodiment, the target digital human business card is provided with an interactive trigger area, which may be a button, a text box or any other form of input interface, and based on the interactive trigger area, the query instruction input by the user may be received.

[0029] Step S30, determining whether the inquiry instruction has a corresponding standard question.

[0030] The standard questions in this embodiment refer to a series of questions pre-configured by the job seeker corresponding to the target digital human business card.

[0031] In some embodiments, this series of questions is stored separately in a question library. When the electronic device receives a user's query instruction, it compares the query instruction with the questions in the preset question library by using a matching algorithm (such as keyword matching, natural language processing technology, etc.) to determine whether there are standard questions corresponding to the query instruction. If a matching question is found, it is considered that the query instruction has a corresponding standard question. If no matching question is found, it is considered that the query instruction does not have a corresponding standard question.

[0032] In some embodiments, when the electronic device receives a user's query instruction, it checks whether the query instruction meets the characteristics of a certain rule by calling the rule engine to determine whether there is a standard question corresponding to the query instruction. If a matching rule is found, it is considered that there is a corresponding standard question for the query instruction. If no matching rule is found, it is considered that there is no corresponding standard question for the query instruction. Among them, a set of rules are defined in the rule engine, and each rule describes the characteristics of a question (such as keywords, sentence structure, etc.).

[0033] In some embodiments, Figure 2 As shown, the step S30 includes: Step S31, obtaining an information set associated with the target digital human business card, wherein the information set includes a basic information set and a question-and-answer information set.

[0034] In this embodiment, the basic information set includes the basic information of the job seeker corresponding to the target digital person's business card, such as user ID, headshot, personal information, information about companies that have worked for, etc. Compared with the content covered in the resume or text introduction area, the content included in the basic information set can be richer, more comprehensive and detailed. The question and answer information set includes a set of preset pre-configured questions and corresponding answers, and can also include historical chat records of the job seeker corresponding to the target digital person's business card during the job search.

[0035] Step S32, calculating the semantic similarity between the query instruction and each question in the question set; In this embodiment, natural language processing technology (such as word embedding, sentence embedding, etc.) is used to perform semantic analysis on the query instruction and each question in the question set. The semantic similarity between the query instruction and each question is calculated. The semantic similarity in this embodiment can be cosine similarity, Jaccard similarity, etc.

[0036] Step S33, comparing the calculated semantic similarity with the preset similarity to determine whether the query instruction has a corresponding standard question, wherein the question with a semantic similarity greater than the preset similarity is the standard question corresponding to the query instruction.

[0037] In this embodiment, a similarity is preset to determine whether the query instruction is similar enough to the questions in the question set.

[0038] For example, suppose that there are Question 1: How can I contact you? Question 2: I want to know the reason for your job change in the question and answer set. The received inquiry instruction is: I want to know how to contact you. The semantic similarity between "I want to know how to contact you." and Question 1 and Question 2 is calculated respectively. Assume that the semantic similarity with Question 1 is 0.95, the semantic similarity with Question 2 is 0.65, and the preset similarity is 0.9. 0.95>0.9>0.65, then Question 1 is determined to be the standard question corresponding to the inquiry instruction.

[0039] Step S40: If there is a standard question corresponding to the inquiry instruction, the standard question and the information set associated with the target digital human business card are input into the pre-trained information desensitization model to obtain the desensitized answer output by the information desensitization model.

[0040] In this embodiment, the standard question and the information set associated with the target digital person's business card are input into the pre-trained information desensitization model, and the pre-trained information desensitization model calls the information set to generate a desensitized answer to the standard question.

[0041] It should be understood that the information desensitization model in this embodiment is a pre-trained large language model used to process and generate desensitized text answers. The model is designed to protect the sensitive information of job seekers while providing useful and accurate answers.

[0042] For example, assuming that the inquiry instruction received by the electronic device based on the target digital person's business card is "What are your main responsibilities in your previous company?", the standard question corresponding to the inquiry instruction is determined to be "What are your main work contents in your previous company?". The electronic device inputs the standard question and the information set associated with the target digital person's business card into the pre-trained information desensitization model. Based on the possible content contained in this information set, such as "Worked as a software engineer at ABC Company, responsible for developing and maintaining the company's core product modules, and participated in the development of multiple large projects", the desensitized answer output by the information desensitization model may be "Worked as a software engineer at the previous company, responsible for developing and maintaining the company's core product modules, and participated in the development of multiple large projects.".

[0043] Step S50: If there is no standard question corresponding to the query instruction, call the preset search engine to search the query instruction, and determine the search data set based on the search results, and input the query instruction, the target digital human business card and the search data set into the pre-trained information desensitization model to obtain the desensitized answer output by the information desensitization model.

[0044] In this embodiment, if there is no standard question corresponding to the query instruction, a preset search engine is called to search the query instruction. The preset search engine includes a general search engine (applicable to questions in a wide range of fields) and / or a search engine in a specific field (applicable to questions in a specific field, such as academic research, industry information, etc.). Collect the first N search results returned by the preset search engine. Usually, the value of N can be set according to actual needs, such as the first 20 or the first 50 results. Evaluate the search results according to the preset evaluation strategy, and select the M search results with the highest comprehensive evaluation scores as the search data set. Input the query instruction, the target digital human business card and the search data set into the pre-trained information desensitization model to obtain the desensitized answer output by the information desensitization model.

[0045] In some embodiments, the electronic device determines the question type of the query instruction by acquiring and analyzing the semantic parsing result of the query instruction, determines a preset search engine to be called according to the question type, and calls the preset search engine to be called to search for the query instruction.

[0046] In some implementations, the content requirement type may also be determined based on the question type. The content requirement type may be "specific experience", "skill certification", "project case", etc. For example, for a question of the "work experience" type, the content requirement type may be "specific experience". The original source of the content may be determined based on the content requirement type. The original source may be a job seeker's peer work experience exchange website, etc. Based on the original source of the content, the preset search engine to be called is determined.

[0047] In some embodiments, the electronic device performs preliminary screening of search results based on preliminary screening indicators to obtain an initial search data set, wherein the preliminary screening indicators include relevance evaluation indicators and / or source credibility evaluation indicators; based on deep screening indicators, the initial search data set is deeply screened to obtain a target search data set, wherein the deep screening indicators include content quality evaluation indicators and repetition rate indicators. Among them, the content quality evaluation indicator is used to evaluate the content quality of the search results, such as whether specific useful information and examples are provided. The repetition rate indicator is used to remove duplicate or highly similar search results and retain only the most representative and diverse content.

[0048] In some embodiments, the search results are preliminarily screened based on the relevance evaluation index to obtain an initial search set. Specifically, the relevance of each search result may be evaluated by at least one of the methods such as keyword matching, title similarity, and abstract content, and the search results with the highest relevance are placed in the initial search data set. For example, if the query instruction is "What experience do you have in project management?", the initial search set containing relevant words such as "project management" and "experience" is screened out.

[0049] In some implementations, the search results are preliminarily screened based on the source credibility evaluation index to obtain an initial search set. Specifically, the search results may be preliminarily screened based on the source of the search results to determine that the search results from authoritative websites, well-known forums, professional blogs, etc. meet the source credibility evaluation index. The search results that meet the source credibility evaluation index are placed in the initial search data set.

[0050] Step S60, obtaining the desensitized answer output by the information desensitization model, and outputting the desensitized answer.

[0051] In this embodiment, the desensitized answer output by the information desensitization model is obtained, and the generated desensitized answer is output to the terminal that issues the inquiry instruction through an interface display or voice broadcast.

[0052] In this embodiment, on the one hand, the digital business card of the present application can be integrated into a variety of recruitment platforms and social media to achieve multi-channel synchronization of information and improve the efficiency of job information dissemination. On the second hand, the digital business card of the present application can provide a more personalized display method, which is convenient for the recruiter to have a more intuitive understanding of the job seeker, and improves the efficiency and accuracy of the transmission of the job seeker's personal information. On the third hand, the digital business card of the present application can respond to the recruiter's inquiry instructions immediately without waiting for the job seeker's reply, reducing the time cost of communication. On the fourth hand, for preset standard questions, the present application can directly call the pre-trained information desensitization model to quickly generate answers, avoid the situation of repeatedly answering the same questions for job seekers, and save the time and energy of job seekers. On the fifth hand, for non-standard questions, the present application can obtain relevant information through a search engine, and generate answers in combination with a pre-trained large model, ensuring the flexibility and wide applicability of the application of the present application. On the sixth hand, the answers output by the present application are desensitized answers, which can provide accurate and useful information to the recruiter under the premise of protecting the privacy of job seekers, further improving the interactive efficiency of the recruitment process.

[0053] like Figure 3 As shown, based on any of the above implementations, before step S10, the step further includes: Step S00, receiving a target digital person business card generation instruction, wherein the target digital person business card generation instruction includes the style type and resume information of the target digital person business card, and the style type of the target digital person business card includes the target digital person style type and the business card typesetting style type.

[0054] In this embodiment, the target digital human style types may include cartoon style, realistic style, business style, creative style, etc. The business card layout style types may include simple style, business style, custom style, etc.

[0055] Step S01, inputting the resume information into a pre-trained information extraction model, wherein the pre-trained information extraction model extracts and outputs the portrait information and standard information in the resume information.

[0056] The standard information in this embodiment refers to the key data that conforms to a specific format and specification extracted from the resume information by the information extraction model. After being standardized, these data can be recognized and used by electronic devices to generate and display important content on the target digital person's business card. Standard information usually includes but is not limited to basic personal information, professional background, educational experience, skill certificates, etc. They are presented in a structured form to facilitate subsequent automated processing and display. Step S02: generating a target digital human image based on the target digital human style type and the portrait information.

[0057] In this embodiment, based on the target digital human style type and the portrait information, the target digital human image is generated by image generation technology (such as GANs) or image editing technology. Among them, the image generation technology can generate a target digital human image that meets the requirements according to the specified target digital human style type. The image editing technology can edit and optimize the existing portrait information to make it meet the specified target digital human style type.

[0058] Step S03, generating and outputting a target digital human business card based on the target digital human image, the standard information and the business card typesetting style type.

[0059] In this embodiment, a target digital person business card is generated based on the target digital person image, the standard information and the business card typesetting style, and the target digital person business card is output to the terminal that issues the inquiry instruction through an interface display.

[0060] In this embodiment, firstly, job seekers can select the style type of the target digital person's business card according to their preferences, including the target digital person's style type (such as cartoon style, realistic style, etc.) and the business card layout style type (such as simple style, business style, etc.), thereby generating a personalized digital person's business card. Secondly, job seekers only need to provide resume information and select the style type, and the electronic device can automatically generate a business card, reducing the complexity of manual operation. Thirdly, the pre-trained information extraction model can quickly and accurately extract portrait information and standard information (such as name, position, contact information, etc.) from the resume information, improving the efficiency and accuracy of information processing.

[0061] like Figure 4 As shown, based on any of the above implementations, before step S30 or step S40, it includes: Step A10, obtaining an initial training data set, wherein the initial training data set includes initial training data of job seekers of various industries, various positions and various experience levels, and the initial training data includes resume information data and dialogue data.

[0062] In this embodiment, the initial training data set is the original data set used to train the information desensitization model, which contains resume information data and conversation data of job-seeking roles in various industries, positions and experience levels. The resume information data contains detailed background information of the job-seeking role, including educational background, work experience, skills and other relevant information. The conversation data can be a real conversation record between the job-seeking role and the recruiter, or it can be the role conversation data constructed by the technicians of the training model based on the job-seeking role’s field and work experience (junior, intermediate, senior, etc.). For example, the recruiter asked: "Please introduce the project you were responsible for in your previous job." The job-seeking role replied: "In my previous job, I was responsible for the development of a large e-commerce platform, mainly using Java and Spring frameworks.".

[0063] Step A20, input the resume information, the conversation data and the instruction data into the information desensitization model to train the information desensitization model to output answer data to the instruction data based on the resume information and the conversation data.

[0064] In this embodiment, the instruction data refers to a specific question instruction, which is used to instruct the information desensitization model to generate a corresponding answer based on the resume information and the conversation data. For example, the instruction data may be "Please describe your project management experience." The answer data generated by the information desensitization model may be "I have 2 years of project management experience and have led a team of 5 people to complete a medium-sized software development project."

[0065] In some implementations, sensitive information in resume information and conversation data may be labeled first; the resume information and conversation data labeled with sensitive information may be input into the information desensitizing model to train the information desensitizing model to output a model response to the instruction data based on the resume information and the conversation data, wherein the sensitive information in the model response may be replaced with desensitized information.

[0066] Specifically, sensitive information may include but is not limited to addresses, telephone numbers, ID numbers, bank account information, information involving trade secrets, etc. The job seeker may not realize that this information is sensitive information, so it appears in the resume information or conversation data, and electronic devices are required to mark this sensitive information for subsequent processing. Next, the resume information and conversation data marked with sensitive information are input into the information desensitization model. This large model has been pre-trained and fine-tuned to recognize and process sensitive information. By inputting the labeled data, the model can learn how to automatically replace sensitive information with desensitized information when generating answers.

[0067] For example, if there is a sentence in the conversation data, "My address is XX Community, Chaoyang District, Beijing." Before entering the model, this information is annotated. When the model generates an answer, it will replace "XX Community, Chaoyang District, Beijing" with a common desensitized identifier or a large-scale address, such as "[Add]" or "China".

[0068] This implementation not only protects the privacy of users, but also ensures that the answers generated by the model comply with data protection regulations and privacy policies. In this way, the model can demonstrate higher security and reliability when processing sensitive information. In addition, the model's answers can be further fine-tuned and optimized to make them more natural and in line with human language habits, while ensuring that sensitive information is effectively desensitized.

[0069] Step A30, obtaining the standard answer corresponding to the instruction data, and calculating the answer loss based on the answer data and the standard answer.

[0070] In this embodiment, the standard answer refers to a pre-set standard answer to the instruction data. The answer loss refers to the difference between the model output answer data and the standard answer quantified by a loss function (such as cross entropy loss, mean square error loss, etc.).

[0071] For example, the instruction data may be "Please describe your project management experience". The answer data generated by the information masking model may be "I have 2 years of project management experience and have led a team of 5 people to complete a medium-sized software development project." The standard answer may be "I have 2 years of project management experience and have successfully led a team of 5 people to successfully complete a medium-sized software development project."

[0072] Step A40, based on the answer loss, the information desensitization model is feedback updated by elastic weight sharing until the training is completed.

[0073] In this embodiment, elastic weight sharing is a model optimization technology that can reduce the complexity and overfitting risk of the model by sharing some weights at different levels or positions while maintaining the performance of the model. Based on the calculated answer loss, the model parameters of the information desensitization model are updated through the back propagation algorithm. This process is iterated continuously until the answer loss reaches a predetermined threshold or the number of training rounds ends.

[0074] In this embodiment, on the one hand, by collecting initial training data from job seekers in various industries, various positions, and various experience levels, the model can process diverse data and improve the generalization ability of the model. On the other hand, by annotating sensitive information in resume information and conversation data, it is ensured that the model can recognize such sensitive information. At the same time, sensitive information in the generated answer data is replaced with desensitized information, effectively protecting the privacy and security of users. On the third hand, through elastic weight sharing, the model can share weights between different tasks or data sets while retaining optimization space for specific tasks, thereby improving the adaptability and performance of the model in different scenarios. On the fourth hand, by calculating the answer loss and performing feedback updates, the model can be continuously optimized to improve the accuracy and quality of the answers.

[0075] like Figure 5 As shown, based on any of the above implementations, after step A40, the following steps are included: Step A50, constructing a reinforcement learning training data set, wherein the reinforcement learning training data set includes the instruction data, the model answer corresponding to the instruction data, the standard answer corresponding to the instruction data, and the answer evaluation index.

[0076] In this embodiment, the model answer refers to the answer corresponding to the instruction data output by the information desensitization model trained by the above implementation method. The answer evaluation index is an index used to evaluate the quality of the model answer, such as accuracy, fluency, desensitization effect, etc.

[0077] Step A60, based on the reinforcement learning training data set, set up a reinforcement learning environment, and define a state space, an action space and a reward function in the reinforcement learning environment, wherein the state space includes instruction data and a model answer corresponding to the instruction data, the action space includes an answer adjustment strategy, and the reward function is used to calculate a reward signal based on a standard answer corresponding to the instruction data and the answer evaluation index.

[0078] In this embodiment, the reinforcement learning environment is a simulated real-world environment for training models. The state space includes all possible sets of states in the reinforcement learning environment. In this embodiment, the state includes instruction data and the model answer corresponding to the instruction data. The action space includes all possible sets of actions in the reinforcement learning environment. In this embodiment, the action includes answer adjustment strategies, such as increasing the degree of desensitization, improving the fluency of the answer, modifying the wording of the answer, adding more details, simplifying the language, etc. The reward function is used to evaluate the effect of the model taking an action in a specific state and return a numerical reward signal. The higher the reward signal, the better the model performs.

[0079] Step A70: In the reinforcement learning environment, the information desensitization model is updated based on the reward signal until the reinforcement learning training is completed.

[0080] In this embodiment, the parameters of the information desensitization model are adjusted according to the reward signal in the reinforcement learning environment to optimize the performance of the model. When the performance of the model reaches a predetermined performance indicator or the number of training times reaches an upper limit, the training is stopped.

[0081] In some embodiments, Figure 6 As shown, the step A70 includes: Step A71, based on the specified instruction data in the state space and the model answer corresponding to the specified instruction data, select an answer adjustment strategy in the action space.

[0082] In this embodiment, the action space includes all possible answer adjustment strategies, such as increasing the degree of desensitization, improving the fluency of the answer, correcting erroneous information, etc. The information desensitization model selects a suitable answer adjustment strategy based on the current state (specified instruction data and the model answer corresponding to the specified instruction data) through a greedy algorithm, random selection, or based on historical performance. Selecting an answer adjustment strategy based on historical performance means that the trained information desensitization model may not perform well on the degree of desensitization of the answer and fail to completely desensitize certain key information. Therefore, an answer adjustment strategy that increases the degree of desensitization is selected.

[0083] Step A72, applying the selected answer adjustment strategy to the trained information desensitization model to generate a new model answer corresponding to the specified instruction data.

[0084] This means that the model answer corresponding to the specified instruction data is adjusted using the selected answer adjustment strategy to generate a new model answer.

[0085] Step A73, calculating a reward signal based on the new model answer corresponding to the specified instruction data, the standard answer corresponding to the specified instruction data and the reward function.

[0086] In this embodiment, the reward function is used to evaluate the difference between the new model answer and the standard answer and generate a reward signal. The reward signal can be a numerical value that indicates the quality of the answer. For example, if the new model answer is very close to the standard answer, the reward signal may be high; otherwise, it is low. The reward signal is calculated by the reward function and is used to guide the subsequent optimization of the model.

[0087] Step A74: based on the reward signal, update the information desensitization model.

[0088] In this embodiment, the model parameters are adjusted according to the calculated reward signal to optimize its performance. The purpose is to enable the model to generate results that are closer to the standard answer in future predictions. When the performance of the model reaches a predetermined performance indicator or the number of training times reaches an upper limit, the training is stopped.

[0089] In this embodiment, the trained information desensitization model is further trained through reinforcement learning, and the performance of the information desensitization model is gradually optimized by continuously selecting answer adjustment strategies, generating new answers, calculating reward signals, and updating the model. This ensures that the model can generate higher quality answers when processing various instruction data.

[0090] Based on any of the above implementation modes, the reinforcement learning training data set constructed in step A50 is divided into multiple subsets; a corresponding reinforcement learning environment is created for each subset, wherein each reinforcement learning environment corresponds to a training process; in each reinforcement learning environment, the gradient information of each training process is calculated; and the information desensitization model is updated by aggregating the gradient information of all training processes.

[0091] In this embodiment, by dividing the data set into multiple subsets and creating an independent reinforcement learning environment for each subset, parallel training is achieved, which can make full use of computing resources. By calculating and summarizing gradient information, the model can be effectively updated to improve training efficiency and model performance. This method is particularly suitable for training large-scale data sets and complex models, and can significantly shorten the training time and improve the convergence speed of the model.

[0092] In one embodiment, a storage medium is provided, on which executable instructions are stored. When the instructions are executed by a processor, the processor executes the steps in the above-mentioned method embodiments.

[0093] In one embodiment, an electronic device is also provided, comprising one or more processors; a memory, wherein one or more programs are stored in the memory, wherein when the one or more programs are executed by one or more processors, the one or more processors execute the steps in the above-mentioned method embodiments.

[0094] In one embodiment, Figure 7, which shows a schematic diagram of the structure of an electronic device for implementing an embodiment of the present application. The electronic device 700 includes a central processing unit (CPU) 701, which can perform various appropriate actions and processes based on a program stored in a read-only memory (ROM) 702 or a program loaded from a storage part 708 to a random access memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the electronic device 700 are also stored. The CPU 701, the ROM 702, and the RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0095] The following components are connected to the I / O interface 705: an input section 706 including a keyboard, a mouse, etc.; an output section 707 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN card, a modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the I / O interface 705 as needed. A removable medium 711, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 710 as needed, so that a computer program read therefrom is installed into the storage section 708 as needed.

[0096] In particular, based on the embodiments of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments of the present application include a computer program product, including a computer-readable medium carrying instructions, in such an embodiment, the instructions can be downloaded and installed from a network through the communication part 709, and / or installed from a removable medium 711. When the instructions are executed by the central processing unit (CPU) 701, the various method steps described in the present application are executed.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

[0098] In addition, those skilled in the art will appreciate that, although some embodiments herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present application and form different embodiments. For example, in the above claims, any one of the claimed embodiments may be used in any combination. The information disclosed in this background technology section is intended only to deepen the understanding of the overall background technology of the present application and should not be regarded as an admission or in any form of implication that the information constitutes prior art known to those skilled in the art.

Claims

1. A digital human business card interaction method, characterized in that: The method comprises: In response to a business card viewing instruction, outputting a target digital person business card associated with the business card viewing instruction, wherein the target digital person business card includes an image display area and a text introduction area; Receiving an inquiry instruction based on the target digital human business card, and determining whether the inquiry instruction has a corresponding standard question; If there is a standard question corresponding to the inquiry instruction, input the standard question and the information set associated with the target digital person business card into the pre-trained information desensitization model to obtain a desensitized answer output by the information desensitization model; If there is no standard question corresponding to the query instruction, calling a preset search engine to search the query instruction, and determining a search data set based on the search results, inputting the query instruction, the target digital person business card and the search data set into a pre-trained information desensitization model to obtain a desensitized answer output by the information desensitization model; Obtain the desensitized answer output by the information desensitization model, and output the desensitized answer.

2. The interactive method of the digital human business card according to claim 1 is characterized in that: The determining whether the inquiry instruction has a corresponding standard question includes: Acquire an information set associated with the target digital person's business card, the information set including a basic information set and a question-and-answer information set; Calculating the semantic similarity between the query instruction and each question in the question set; The calculated semantic similarity is compared with the preset similarity to determine whether the query instruction has a corresponding standard question, wherein the question with a semantic similarity greater than the preset similarity is the standard question corresponding to the query instruction.

3. The interactive method of the digital human business card according to claim 2 is characterized in that: Before outputting the target digital person business card associated with the business card viewing instruction in response to the business card viewing instruction, the method includes: Receive a target digital person business card generation instruction, the target digital person business card generation instruction includes the target digital person business card style type and resume information, the target digital person business card style type includes the target digital person style type and business card layout style type; Inputting the resume information into a pre-trained information extraction model, wherein the pre-trained information extraction model extracts and outputs the portrait information and standard information in the resume information; Based on the style type of the target digital human and the portrait information, generate a target digital human image; Based on the target digital human image, the standard information and the business card typesetting style type, a target digital human business card is generated and output.

4. The interactive method of the digital human business card according to claim 1 is characterized in that: Before inputting the information set associated with the standard question and the target digital person's business card into the pre-trained information desensitization model, or inputting the query instruction, the target digital person's business card and the search data set into the pre-trained information desensitization model, the method includes: Acquire an initial training data set, wherein the initial training data set includes initial training data of job seekers of various industries, various positions, and various experience levels, and the initial training data includes resume information data and conversation data; Inputting the resume information, the conversation data and the instruction data into the information desensitization big model to train the information desensitization big model to output answer data to the instruction data based on the resume information and the conversation data; Obtaining a standard answer corresponding to the instruction data, and calculating an answer loss based on the answer data and the standard answer; Based on the answer loss, the information desensitization model is feedback updated by using elastic weight sharing until the training is completed.

5. The interactive method of the digital human business card according to claim 4 is characterized in that: The step of inputting the resume information, the conversation data and the instruction data into the information desensitization big model to train the information desensitization big model to output a model answer to the instruction data based on the resume information and the conversation data includes: Marking sensitive information in the resume information and the conversation data; The resume information and conversation data marked with sensitive information are input into the information desensitizing model to train the information desensitizing model to output a model response to the instruction data based on the resume information and the conversation data, wherein the sensitive information in the model response will be replaced with desensitized information.

6. The interactive method of the digital human business card according to claim 4 is characterized in that: Based on the answer loss, the information desensitization model is fed back and updated in an elastic weight sharing manner until the training is completed, including: Constructing a reinforcement learning training data set, wherein the reinforcement learning training data set includes the instruction data, a model answer corresponding to the instruction data, a standard answer corresponding to the instruction data, and an answer evaluation index; Based on the reinforcement learning training data set, a reinforcement learning environment is set, and a state space, an action space and a reward function are defined in the reinforcement learning environment, wherein the state space includes instruction data and a model answer corresponding to the instruction data, the action space includes an answer adjustment strategy, and the reward function is used to calculate a reward signal based on a standard answer corresponding to the instruction data and the answer evaluation index; In the reinforcement learning environment, the information desensitization model is updated based on the reward signal until the reinforcement learning training is completed.

7. The interactive method of the digital human business card according to claim 6 is characterized in that: In the reinforcement learning environment, updating the information desensitization model based on the reward signal includes: Based on the specified instruction data in the state space and the model answer corresponding to the specified instruction data, selecting an answer adjustment strategy in the action space; Applying the selected answer adjustment strategy to the trained information desensitization model to generate a new model answer corresponding to the specified instruction data; Calculating a reward signal based on the new model answer corresponding to the specified instruction data, the standard answer corresponding to the specified instruction data and the reward function; Based on the reward signal, the information desensitization model is updated.

8. The interactive method of the digital human business card according to claim 6 is characterized in that: The step of constructing a reinforcement learning training data set includes: Constructing a reinforcement learning training data set, and dividing the reinforcement learning training data set into multiple subsets; The step of setting a reinforcement learning environment based on the reinforcement learning training data set includes: Create a corresponding reinforcement learning environment for each subset, where each reinforcement learning environment corresponds to a training process; In the reinforcement learning environment, updating the information desensitization model based on the reward signal includes: In each reinforcement learning environment, calculate the gradient information of each training process; The information desensitization model is updated by aggregating the gradient information of all training processes.

9. An electronic device, characterized in that: include: one or more processors; a memory for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors execute the digital human business card interaction method as claimed in any one of claims 1 to 8.

10. A storage medium, characterized in that: The storage medium stores executable instructions, and when the instructions are executed by the processor, the processor executes the digital human business card interaction method according to any one of claims 1 to 8.

Images