Conversational occupational interest dynamic correction method and device, equipment and storage medium

Abstract

The invention relates to the technical field of information processing, in particular to a dialogue type occupational interest dynamic correction method, device and equipment and a storage medium, by collecting preliminary occupational willingness information of a student, the student is guided to complete Holland occupational interest evaluation based on a dialogue type Agent, a multi-dimensional user portrait is constructed according to an occupational interest evaluation result, and the user experience is improved. And comparing the occupational interest evaluation result with the preliminary occupational willingness information, identifying the difference in the comparison result and generating a difference tag, performing multi-round interaction with students through a dialogue Agent based on the difference tag, and dynamically correcting the multi-dimensional user portrait according to the multi-round interaction information, thereby significantly improving the authenticity of the user portrait, and improving the user experience. The method comprises the steps of constructing a post knowledge graph based on a convergence portrait, generating post recommendation information based on the post knowledge graph in combination with an interest matching degree, a skill matching degree and constraint compatibility, and outputting an interpretable recommendation reason, so that the transparency of a system is enhanced, and the satisfaction degree of a user is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information processing, and in particular to a dialogue type professional interest dynamic correction method, device, equipment and storage medium. BACKGROUND

[0002] With the advent of the digital and artificial intelligence era, the speed of traditional occupation extinction and emerging occupation emergence is accelerating, making the uncertainty of individual career development increasing. Under this background, many college students are generally faced with confusion and difficulties when facing the direction of graduation. As an important theoretical basis for career recommendation and personnel evaluation, the Person-Job Fit (P-J Fit) theory is derived from the Person-Environment Fit (P-E Fit) theory. Under this framework, the attitude, behavior and performance of individuals are not determined by personal qualities or environment alone, but by the degree of fit between the two. A large number of studies have shown that when individual characteristics and work environment are highly compatible, higher job satisfaction and performance will be obtained, otherwise, it will lead to stress, burnout and even performance decline.

[0003] In recent years, the development of AI technology has provided new tools for career recommendation and person-job fit. For example, Wang et al. (2022) proposed a "recruitment history enhanced matching mechanism", which uses the historical interaction (such as browsing, applying, and interviewing) between candidates and positions to build a co-attention neural network (Co-Attention NN), significantly improving the semantic alignment effect of matching. Gong et al. (2024) further proposed the WEPJM model, which incorporates the three dimensions of "consistency", "similarity" and "continuity" in the career path into modeling, significantly improving the accuracy of high-level job recommendation in the multi-task and contrast learning framework. However, the existing research generally has a commonly overlooked premise assumption: "people" are equivalent to "resumes", that is, it is assumed that job seekers can clearly and accurately express their interests, skills and goals. However, for college students who are still in the exploration stage, this assumption does not hold, because their expressions of interest preferences, skill evidence and career expectations are often one-sided or even contradictory. At the same time, the job side description has problems such as inconsistent terminology, inconsistent granularity and high noise, and lacks verification and correction of the authenticity and consistency of self-reported information, which weakens the effectiveness of traditional static matching, leading to large errors in job recommendation results and reducing the reliability of career guidance systems and user satisfaction. SUMMARY

[0004] In order to overcome the deficiencies of the prior art, the purpose of the present application is to provide a dialogue type career interest dynamic correction method, device, equipment and storage medium which can verify and correct the authenticity and consistency of user self-reported information, significantly improve the authenticity of user portraits, prevent errors caused by self-reported deviations, improve the accuracy of job recommendations, generate traceable recommendation reasons, enhance the transparency of the system, and improve user satisfaction.

[0005] The first aspect of the present application provides a dialogue type career interest dynamic correction method, comprising: collecting preliminary career intention information of a student, guiding the student to complete a Holland career interest evaluation based on a dialogue type Agent, and obtaining a career interest evaluation result; constructing a multi-dimensional user portrait according to the career interest evaluation result, comparing the career interest evaluation result with the preliminary career intention information, and obtaining a comparison result; identifying differences in the comparison result and generating a difference label, based on the difference label, performing multi-round interaction with the student through a dialogue type Agent to explain the source of the differences and guide the student to clarify the real intention, and obtaining multi-round interaction information; dynamically correcting the multi-dimensional user portrait according to the multi-round interaction information, obtaining a convergent portrait, constructing a job knowledge graph based on the convergent portrait; based on the job knowledge graph, combining interest matching degree, skill matching degree and constraint compatibility to generate job recommendation information, and outputting an interpretable recommendation reason.

[0006] Optionally, in the first implementation manner of the first aspect of the present application, the collecting of the preliminary career intention information of the student, the guiding of the student to complete the Holland career interest evaluation based on the dialogue type Agent, and the obtaining of the career interest evaluation result, comprise: collecting the preliminary career intention information of the student through a natural language interaction interface, the preliminary career intention information including career preference, skill self-evaluation and career values; using the dialogue type Agent to generate interactive questions based on the Holland career interest evaluation scale, and guiding the student to complete the interactive questions through multi-round dialogue to obtain the answer information of the student; using natural language processing technology to analyze the answer information and calculate Holland six-type interest scores to obtain the career interest evaluation result.

[0007] Optionally, in the second implementation form of the first aspect of the present application, the constructing a multi-dimensional user portrait according to the career interest assessment result, and comparing the career interest assessment result with the preliminary career aspiration information to obtain a comparison result, comprises: extracting interest dimension features based on the career interest assessment result, and constructing a multi-dimensional user portrait according to the interest dimension features and the personal basic information of the student; performing semantic alignment on the career interest assessment result and the preliminary career aspiration information to obtain an alignment result, and calculating an interest consistency score according to the alignment result; when the interest consistency score is lower than a preset score threshold, generating a comparison result, the comparison result comprising consistent items, inconsistent items and a difference degree.

[0008] Optionally, in the third implementation form of the first aspect of the present application, the identifying the difference in the comparison result and generating a difference label, and based on the difference label, performing multi-round interaction with the student through a conversational Agent to explain the source of the difference and guide the student to clarify the real aspiration to obtain multi-round interaction information, comprises: identifying the difference type in the comparison result using a rule engine and a machine learning model, the difference type being one or more of interest preference conflict, skill cognitive bias and insufficient career information; generating a difference label for each of the difference types, and constructing a difference explanation template; performing multi-round interaction with the student through the conversational Agent based on the difference label and the difference explanation template to provide difference source analysis and guide the student to clarify the real aspiration; recording the conversation content of the multi-round interaction, and extracting key clarification information in the conversation content as multi-round interaction information.

[0009] Optionally, in the fourth implementation form of the first aspect of the present application, the dynamically correcting the multi-dimensional user portrait according to the multi-round interaction information to obtain a convergent portrait, and constructing a job knowledge graph based on the convergent portrait, comprises: updating the interest weight and the skill label in the multi-dimensional user portrait according to the multi-round interaction information using an incremental learning algorithm to obtain a convergent portrait; extracting job entities, skill requirements and interest matching conditions from public job descriptions of a job database; constructing a job knowledge graph according to the job entities, the skill requirements and the interest matching conditions, the job knowledge graph comprising job nodes, skill nodes, interest nodes and relationship edges; writing the interest information, the ability information and the constraint information in the convergent portrait into the job knowledge graph; and converting the job knowledge graph into a semantic vector using a graph embedding technology.

[0010] Optionally, in the fifth implementation form of the first aspect of the present application, the post recommendation information is generated based on the post knowledge graph in combination with the interest matching degree, the skill matching degree and the constraint compatibility, and an interpretable recommendation reason is output, including: calculating the interest matching degree between the interest information in the converged portrait and the post node in the post knowledge graph; calculating the skill matching degree between the ability information in the converged portrait and the post node in the post knowledge graph; evaluating the constraint compatibility between the constraint information in the converged portrait and the post node in the post knowledge graph, the constraint compatibility including geographical location compliance degree and salary expectation compliance degree; calculating the interest matching degree, the skill matching degree and the constraint compatibility based on a preset weight ratio to obtain a comprehensive recommendation score; sorting a plurality of candidate posts according to the comprehensive recommendation score to obtain candidate post sorting information, and generating a post recommendation list as the post recommendation information based on the candidate post sorting information; and generating an interpretable recommendation reason based on the post recommendation information, the recommendation reason including interest matching explanation, skill gap analysis and constraint satisfaction situation.

[0011] Optionally, in the sixth implementation form of the first aspect of the present application, after the post recommendation information is generated based on the post knowledge graph in combination with the interest matching degree, the skill matching degree and the constraint compatibility, and the interpretable recommendation reason is output, the method further includes: sending the post recommendation information and the recommendation reason to the user terminal corresponding to the student, so that the user terminal generates and displays a visual recommendation page; receiving feedback information of the user terminal, the feedback information including recommendation satisfaction and user's final post selection; and adjusting the weight ratio used for the weighted fusion calculation of the interest matching degree, the skill matching degree and the constraint compatibility according to the feedback information.

[0012] The second aspect of the present application provides a dialogue type occupation interest dynamic correction device, comprising: a collection guiding module, configured to collect preliminary occupation intention information of a student, guide the student to complete a Holland occupation interest evaluation based on a dialogue type agent, and obtain an occupation interest evaluation result; a construction comparison module, configured to construct a multi-dimensional user portrait according to the occupation interest evaluation result, compare the occupation interest evaluation result with the preliminary occupation intention information, and obtain a comparison result; an identification generation interaction module, configured to identify differences in the comparison result and generate a difference label, based on the difference label, perform multi-round interaction with the student through the dialogue type agent to explain the source of the differences and guide the student to clarify the real intention, and obtain multi-round interaction information; a correction construction module, configured to dynamically correct the multi-dimensional user portrait according to the multi-round interaction information, obtain a convergent portrait, and construct a post knowledge graph based on the convergent portrait; and a generation output module, configured to generate post recommendation information based on the post knowledge graph, in combination with interest matching degree, skill matching degree and constraint compatibility, and output an interpretable recommendation reason.

[0013] Optionally, in the first implementation manner of the second aspect of the present application, the collection guiding module comprises: a collection unit, configured to collect preliminary occupation intention information of a student through a natural language interaction interface, the preliminary occupation intention information including occupation preference, skill self-evaluation and occupation values; a generation guiding unit, configured to generate interactive questions based on a Holland occupation interest evaluation scale using a dialogue type agent, and guide the student to complete the interactive questions through multi-round dialogue, to obtain answer information of the student; and an analysis calculation unit, configured to analyze the answer information using a natural language processing technology and calculate Holland six-type interest scores, to obtain an occupation interest evaluation result.

[0014] Optionally, in the second implementation manner of the second aspect of the present application, the construction comparison module comprises: an extraction construction unit, configured to extract interest dimension features based on the occupation interest evaluation result, and construct a multi-dimensional user portrait according to the interest dimension features and personal basic information of the student; an alignment calculation unit, configured to perform semantic alignment on the occupation interest evaluation result and the preliminary occupation intention information, to obtain an alignment result, and calculate an interest consistency score according to the alignment result; and a first generation unit, configured to generate a comparison result when the interest consistency score is lower than a preset score threshold, the comparison result including consistent items, inconsistent items and difference degrees.

[0015] Optionally, in a third implementation form of the second aspect of the present application, the identifying and generating module comprises: an identifying unit configured to identify a difference type in the comparison result using a rule engine and a machine learning model, the difference type being one or more of interest preference conflict, skill cognitive bias, and insufficient professional information; a generating and constructing unit configured to generate a difference label for each of the difference types and construct a difference explanation template; an interacting unit configured to perform multi-round interaction with the student based on the difference label and the difference explanation template through a dialog Agent to provide difference source analysis and guide the student to clarify the real intention; and a recording and extracting unit configured to record a dialogue content of the multi-round interaction and extract key clarification information in the dialogue content as multi-round interaction information.

[0016] Optionally, in a fourth implementation form of the second aspect of the present application, the correcting and constructing module comprises: an updating unit configured to update the interest weight and the skill label in the multi-dimensional user portrait according to the multi-round interaction information using an incremental learning algorithm to obtain a converged portrait; an extracting unit configured to extract a post entity, a skill requirement, and an interest matching condition from a public post description of a professional database; a constructing unit configured to construct a post knowledge graph according to the post entity, the skill requirement, and the interest matching condition, the post knowledge graph comprising a post node, a skill node, an interest node, and a relationship edge; a writing unit configured to write the interest information, the ability information, and the constraint information in the converged portrait into the post knowledge graph; and a converting unit configured to convert the post knowledge graph into a semantic vector using a graph embedding technology.

[0017] Optionally, in a fifth implementation form of the second aspect of the present application, the generating output module comprises: a first calculating unit configured to calculate an interest matching degree between the interest information in the converged portrait and a post node in the post knowledge graph; a second calculating unit configured to calculate a skill matching degree between the ability information in the converged portrait and the post node in the post knowledge graph; an evaluating unit configured to evaluate a constraint compatibility between the constraint information in the converged portrait and the post node in the post knowledge graph, the constraint compatibility comprising a geographical location coincidence degree and a salary expectation coincidence degree; a third calculating unit configured to perform weighted fusion calculation on the interest matching degree, the skill matching degree, and the constraint compatibility based on a preset weight ratio to obtain a comprehensive recommendation score; a sorting generating unit configured to sort a plurality of candidate posts according to the comprehensive recommendation score to obtain candidate post sorting information, and generate a post recommendation list as post recommendation information according to the candidate post sorting information; and a second generating unit configured to generate an interpretable recommendation reason based on the post recommendation information, the recommendation reason comprising interest matching explanation, skill gap analysis, and constraint satisfaction situation.

[0018] Optionally, in a sixth implementation form of the second aspect of the present application, further comprising: a sending module configured to send the post recommendation information and the recommendation reason to the user terminal corresponding to the student, so that the user terminal generates and displays a visual recommendation page; a receiving module configured to receive feedback information of the user terminal, the feedback information including a recommendation satisfaction degree and a final post selection of the user; and an adjusting module configured to adjust a weight proportion used for weighted fusion calculation of the interest matching degree, the skill matching degree and the constraint compatibility according to the feedback information.

[0019] The third aspect of the present application provides a dialog-based dynamic correction device for career interest, comprising a memory and at least one processor, the memory storing instructions; at least one processor calling the instructions in the memory to make the dialog-based dynamic correction device perform each step of the dialog-based dynamic correction method for career interest.

[0020] The fourth aspect of the present application provides a computer readable storage medium, the computer readable storage medium storing instructions, the instructions being executed by a processor to implement each step of the dialog-based dynamic correction method for career interest.

[0021] In the technical solution of the present application, the preliminary career intention information of the student is collected, the student is guided to complete the Holland career interest evaluation based on the dialog-based Agent, the multi-dimensional user portrait is constructed according to the career interest evaluation result, the career interest evaluation result is compared with the preliminary career intention information, the differences in the comparison result are identified and the difference labels are generated, the multi-round interaction between the student and the dialog-based Agent is carried out based on the difference labels, the multi-dimensional user portrait is dynamically corrected according to the multi-round interaction information, the authenticity of the user portrait is significantly improved, the error recommendation caused by the self-reporting bias is prevented, the accuracy of the post recommendation is improved, the post knowledge graph is constructed based on the convergent portrait, the post recommendation information is generated based on the post knowledge graph, combined with the interest matching degree, the skill matching degree and the constraint compatibility, and the explainable recommendation reason is output, the transparency of the system is enhanced, and the user satisfaction is improved. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 The first flowchart of the dialog-based dynamic correction method for career interest provided by the embodiment of the present application; Figure 2 The second flowchart of the dialog-based dynamic correction method for career interest provided by the embodiment of the present application; Figure 3 The third flowchart of the dialog-based dynamic correction method for career interest provided by the embodiment of the present application; Figure 4A fourth flowchart of the dynamic dialogue-type occupation interest correction method provided by the embodiment of the present application is provided. Figure 5 A structural schematic diagram of the dynamic dialogue-type occupation interest correction device provided by the embodiment of the present application is provided. Figure 6 Another structural schematic diagram of the dynamic dialogue-type occupation interest correction device provided by the embodiment of the present application is provided. Figure 7 A structural schematic diagram of the dynamic dialogue-type occupation interest correction device provided by the embodiment of the present application is provided. DETAILED DESCRIPTION

[0023] The present application provides a dynamic dialogue-type occupation interest correction method, device, equipment and storage medium, which can verify and correct the authenticity and consistency of user self-report information, significantly improve the authenticity of user portrait, prevent errors caused by self-report bias, improve the accuracy of post recommendation, can generate traceable recommendation reasons, enhance the transparency of the system, and improve user satisfaction.

[0024] The terms "first", "second", "third", "fourth" and the like in the description, claims, and above drawings of the present application (if any) are used to distinguish similar objects, and do not necessarily have to be used to describe a particular order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" or "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0025] For the sake of understanding, the specific flow of the embodiment of the present application is described below, please refer to Figure 1 One embodiment of the dynamic dialogue-type occupation interest correction method in the embodiment of the present application includes: 101, collecting the preliminary occupation intention information of students, guiding students to complete the Holland occupation interest evaluation based on the dialogue-type Agent, and obtaining the occupation interest evaluation result; In this embodiment, the system collects the preliminary career aspiration information input by the student through a natural language interaction interface (such as a web chat robot or a mobile application), including career preferences (such as "want to become a software engineer"), skill self-evaluation (such as "master Python programming"), and career values (such as "focus on work-life balance"), the conversational Agent generates interactive questions based on the Holland Career Interest Inventory Scale, for example, "do you prefer to solve technical problems alone or in a team?", and guides the student to complete all questions through multiple rounds of dialogue, ensuring a natural and smooth experience, the student's answer information is recorded in real time, and natural language processing technology (such as a BERT-based intent recognition model) is used to analyze the answer content, calculate the Holland six-type interest score (realistic, research, artistic, social, entrepreneurial, and conventional), and finally generate the career interest evaluation result.

[0026] 102. Construct a multi-dimensional user portrait based on the career interest evaluation result, and compare the career interest evaluation result with the preliminary career aspiration information to obtain a comparison result; In this embodiment, the system extracts interest dimension features (such as six-type score distribution) based on the career interest evaluation result, and combines the student's personal basic information (such as major, grade, and internship experience) to construct a multi-dimensional user portrait, which includes interest labels, skill labels, and value labels. Subsequently, the system performs semantic alignment of the career interest evaluation result and the preliminary career aspiration information, calculates the similarity of the two in the semantic space using a word embedding model (such as Word2Vec or Sentence-BERT), obtains an alignment result, calculates an interest consistency score based on the alignment result, and if the score is lower than a preset threshold (such as 0.7), generates a comparison result, including consistent items (such as both the evaluation and the self-report pointing to technical careers), inconsistent items (such as the evaluation showing a high score in social type but the self-report preferring independent work), and the degree of difference (such as mild inconsistency, moderate inconsistency, and severe inconsistency).

[0027] 103. Identify the differences in the comparison result and generate a difference label, based on the difference label, perform multiple rounds of interaction with the student through the conversational Agent to explain the source of the difference and guide the student to clarify the true aspiration, and obtain multiple rounds of interaction information; In this embodiment, the system uses a rule engine (based on predefined logical rules) and machine learning models (such as SVM or neural network classifiers) to identify the types of differences in the comparison results, including interest preference conflicts (such as contradictions between assessment interests and self-reports), skill cognitive biases (such as overestimation or underestimation of skills), and insufficient career information (such as lack of understanding of certain careers by students), generates difference labels (such as "interest conflict - technology vs. social") for each type of difference, and builds difference explanation templates (such as "your assessment shows that you are good at social interaction, but your self-reported preference is for technical work, which may be due to"), and the conversational Agent interacts with the student in multiple rounds based on the difference labels and templates, provides difference source analysis and guides the student to clarify the true intention (such as "can you tell me more about why you prefer technical work?"), and all conversation content is recorded and key clarification information is extracted as multi-round interaction information through key information extraction technology (such as BERT-based QA model).

[0028] 104. Dynamically revise the multi-dimensional user portrait based on the multi-round interaction information to obtain a convergent portrait, and build a job knowledge graph based on the convergent portrait; In this embodiment, the system uses an incremental learning algorithm (such as online gradient descent or incremental K-Means) to dynamically update the interest weights and skill labels in the multi-dimensional user portrait based on the multi-round interaction information, for example, adjust the interest score based on the student's clarification, add new skill labels, to obtain a more accurate convergent portrait, at the same time, the system extracts job entities (such as "Java development engineer"), skill requirements (such as "familiar with Spring framework") and interest matching conditions (such as "suitable for research-oriented personality") from the public job descriptions in the job database (such as recruitment website API or public data set), and builds a job knowledge graph based on these information, where the nodes include job nodes, skill nodes, and interest nodes, and the relationship edges represent the association between jobs and skills, interests, and the interest information, ability information and constraint information (such as geographical location preference) in the convergent portrait are written into the knowledge graph, and the knowledge graph is converted into semantic vectors using graph embedding technology (such as Node2Vec or TransE) to facilitate subsequent matching calculations.

[0029] 105. Based on the job knowledge graph, generate job recommendation information by combining interest matching degree, skill matching degree and constraint compatibility, and output interpretable recommendation reasons; In this embodiment, the system calculates the interest matching degree (such as cosine similarity) of the interest information in the converged image and the post nodes in the post knowledge graph, the skill matching degree (such as Jaccard similarity or cosine similarity) of the ability information and the post nodes, and evaluates the constraint compatibility, including the geographical position coincidence degree (such as the distance between the post location and the preferred city) and the salary expectation coincidence degree (such as whether the post salary meets the expectation), based on the preset weight proportion (such as interest 30%, skill 50%, constraint 20%), the matching degrees are weighted and fused to calculate the comprehensive recommendation score, the candidate posts are sorted according to the score, and the post recommendation list is generated as the post recommendation information, and the system generates an interpretable recommendation reason, including interest matching explanation (such as "the post is suitable for your research interest"), skill gap analysis (such as "you need to strengthen your database skills") and constraint satisfaction (such as "the post is in Beijing, which meets your location preference").

[0030] In the embodiment of the application, the initial multi-dimensional user portrait is constructed after the student's preliminary intention is collected and the standardized Holland evaluation is completed through the natural interactive dialogue Agent, the evaluation results and the self-reported information are innovatively semantically aligned and difference identified, and based on the difference label, a multi-round clarification dialogue is driven, the user portrait is dynamically corrected by using the incremental learning technology, a converged image that more truly reflects the student's internal characteristics is obtained, and finally, based on the multi-dimensional matching algorithm of fused interest, skill and constraint, an interpretable post recommendation is generated on the post knowledge graph, effectively solving the "self-reported bias" problem caused by unclear self-cognition and insufficient information of the student group, and significantly improving the accuracy, reliability and user satisfaction of the career recommendation system.

[0031] Please refer to Figure 2 The second embodiment of the dialogue-based career interest dynamic correction method in the embodiment of the application includes: 201. Collecting the preliminary career intention information of the student through a natural language interaction interface, the preliminary career intention information including career preference, skill self-evaluation and career values; In this embodiment, the system collects the preliminary career intention information input by the student through a chat interface or a form integrated with a natural language processing engine, the career preference is obtained through open-ended questions (such as "What is your ideal job?") or option lists, the skill self-evaluation is collected through a self-evaluation scale (such as "Please rate your programming ability from 1 to 5"), and the career values are collected through multiple-choice or ranking questions (such as "Please rank: salary, development space, work-life balance"), all information is structured and stored for subsequent analysis.

[0032] 202. Generating interactive questions based on the Holland career interest evaluation scale using a dialogue Agent, and guiding the student to complete the interactive questions through multi-round dialogue to obtain the student's answer information; In this embodiment, the conversational Agent dynamically generates a sequence of questions based on the Holland Code, such as "Do you prefer fixing machines or helping others?", and guides the student to complete the assessment through a multi-round dialogue management with context awareness, such as a state machine-based dialogue flow. The question presentation methods include text, options, or scenario simulation to improve engagement, and the student's answer information is recorded in real time, including the original text and selected items.

[0033] 203. Analyze the answer information using natural language processing techniques and calculate the Holland Six-Interest Score to obtain the career interest assessment result; In this embodiment, the system uses natural language processing techniques (such as named entity recognition and sentiment analysis) to analyze the student's answer information. For text answers, an intent classification model (such as a RoBERTa-based text classifier) is used to map to the Holland dimensions, and for option answers, the score is calculated directly according to the scale rules. Finally, the system aggregates the six-type scores to generate the career interest assessment result, including the dominant interest type and score distribution.

[0034] 204. Extract interest dimension features based on the career interest assessment result, and construct a multi-dimensional user portrait based on the interest dimension features and the student's personal basic information; In this embodiment, the system extracts interest dimension features such as the six-type score vector and the dominant type label from the career interest assessment result, and combines the student's personal basic information (such as age, major, and performance) to construct a multi-dimensional user portrait using feature fusion techniques (such as vector concatenation or attention mechanism). The portrait includes static attributes (such as education) and dynamic attributes (such as interest change trend), and is stored in JSON or graph structure.

[0035] 205. Align the career interest assessment result with the preliminary career aspiration information semantically to obtain an alignment result, and calculate an interest consistency score based on the alignment result; In this embodiment, the system uses a semantic alignment model (such as Sentence-BERT) to encode the career interest assessment result (such as "research-oriented high score") and the preliminary career aspiration information (such as "self-reported preference for technical work") into vectors, and calculates the cosine similarity as the alignment result. The interest consistency score is calculated based on the similarity, and if the score is lower than the threshold, the difference detection process is triggered.

[0036] 206. When the interest consistency score is lower than the pre-set score threshold, generate a comparison result, which includes consistent items, inconsistent items, and difference degree; In this embodiment, the system presets a score threshold (e.g., 0.7), and automatically generates comparison results when the interest consistency score is below the threshold. Consistent items and inconsistent items are identified through a rule engine and semantic matching, and the difference degree is classified according to the score deviation and the number of conflicts (e.g., low, medium, and high). The comparison results are output in a structured format for subsequent interaction.

[0037] In the embodiment of the present application, the structured collection of preliminary career intention information is realized through a natural language interaction interface, and the traditional Holland scale is converted into a dynamic and interactive multi-round question and answer by using a conversational Agent, which significantly improves the participation and completion quality of the evaluation. Then, the natural language processing technology is used to accurately analyze the answers and calculate the interest score, laying a data foundation for constructing an initial user portrait. By extracting interest dimension features and fusing them with personal information to construct a multi-dimensional user portrait, the objective evaluation results are deeply semantically aligned and consistent with the subjective self-reported intention, and the inconsistency between the student's internal interest and external expression is automatically and quantitatively detected, providing accurate decision-making basis for subsequent difference analysis and dynamic correction.

[0038] Please refer to Figure 3 The third embodiment of the conversational career interest dynamic correction method in the embodiment of the present application includes: 301. Use a rule engine and a machine learning model to identify the difference types in the comparison results, and the difference types are one or more of interest preference conflict, skill cognitive bias, and insufficient career information; In this embodiment, the system uses a Drools-based rule engine to define difference identification rules (e.g., "If the interest score does not match the self-reported career field, mark it as an interest preference conflict"), and combines a machine learning model (e.g., a random forest or a deep learning classifier) to learn difference patterns from historical data. The difference types are classified as interest preference conflict, skill cognitive bias, and insufficient career information, and are attached with confidence scores.

[0039] 302. Generate a difference label for each difference type and build a difference explanation template; In this embodiment, the system generates a standardized difference label (e.g., "interest conflict - technology vs. art") for each difference type, and builds a difference explanation template based on a template engine (e.g., Jinja2). The template contains dynamic placeholders for inserting specific difference content (e.g., "Your evaluation shows artistic interest, but you prefer technical work, which may be because"), ensuring personalized and natural explanations.

[0040] 303. Through the conversational Agent, multi-round interaction is carried out with the student based on the difference label and the difference explanation template to provide difference source analysis and guide the student to clarify the true intention; In this embodiment, the conversational Agent generates interactive sentences using difference tags and templates, provides difference source analysis through multi-round dialogue (such as dialogue strategy based on reinforcement learning), and guides students to clarify real intentions, for example, the Agent will ask "Can you give an example of your favorite artistic activity?" to collect more information, and the dialogue process is managed based on a state machine to ensure logical coherence.

[0041] 304. Record the dialogue content of the multi-round interaction, and extract the key clarification information in the dialogue content as multi-round interaction information; In this embodiment, the system uses a conversation recording module to store the original dialogue content of the multi-round interaction, including timestamps and speakers, and key clarification information is extracted through information extraction technology (such as a BERT-based sequence labeling model), including clarified interest points, skill descriptions, and career preferences, and is structured and stored as multi-round interaction information.

[0042] 305. Update the interest weight and skill label in the multi-dimensional user portrait using an incremental learning algorithm according to the multi-round interaction information to obtain a converged portrait; In this embodiment, the system applies an incremental learning algorithm (such as an online neural network or incremental clustering) to dynamically adjust the user portrait according to the multi-round interaction information, for example, to update the interest weight (such as increasing the artistic score), add or correct the skill label (such as "Mastering Photoshop"), and obtain a more accurate converged portrait. The portrait update takes effect in real time to ensure data timeliness.

[0043] 306. Extract job entities, skill requirements, and interest matching conditions from public job descriptions in the career database; In this embodiment, the system obtains public job descriptions from a career database (such as LinkedIn or a local recruitment library) through web crawling or API, and uses an information extraction model (such as a BERT-based entity recognition and relationship extraction) to extract job entities (such as "Product Manager"), skill requirements (such as "Requirement Analysis Ability"), and interest matching conditions (such as "Suitable for Enterprise Personality"), and the extraction results are processed for deduplication and standardization.

[0044] 307. Construct a job knowledge graph according to the job entities, skill requirements, and interest matching conditions, the job knowledge graph including job nodes, skill nodes, interest nodes, and relationship edges; In this embodiment, the system uses a graph database (such as Neo4j or JanusGraph) to construct a job knowledge graph, job nodes, skill nodes, and interest nodes are identified by unique IDs, and relationship edges represent "job requires skill", "job matches interest", and other associations. The graph structure supports efficient query and matching calculation.

[0045] 308. Incorporate the interest, ability, and constraint information from the convergence profile into the job knowledge graph; In this embodiment, the system writes the interest information (such as six-type scores), ability information (such as skill tags), and constraint information (such as location preferences) in the converged profile as attributes into the corresponding nodes of the job knowledge graph. For example, the system adds the "interest: research-oriented" attribute to the user node, so that the profile and job data are interconnected.

[0046] 309. Use graph embedding technology to transform job knowledge graphs into semantic vectors; In this embodiment, the system uses graph embedding technology (such as GraphSAGE or DeepWalk) to transform nodes and relationships in the job knowledge graph into low-dimensional semantic vectors. These vectors are used for subsequent similarity calculation and recommendation matching to improve retrieval efficiency and accuracy.

[0047] In this embodiment of the invention, a combination of rule engine and machine learning model is used to intelligently identify and classify the types of differences in profile comparison, and generate standardized difference labels and explanation templates. This provides the core driving force for the conversational agent to conduct targeted and guided clarification interactions. Through multiple rounds of interaction, students are effectively guided to analyze the source of differences, clarify their true intentions, and record and extract key clarification information. Then, the incremental learning algorithm is used to update the user profile in real time and dynamically based on the interaction information, resulting in a converged profile that better reflects the student's true state. Simultaneously, structured information is extracted from the occupational database to construct a job knowledge graph, and the converged profile is integrated into it. Graph embedding technology is used to transform it into semantic vectors, thus building a high-quality, dynamically updatable data foundation and computational framework for subsequent accurate semantic matching and job recommendation.

[0048] Please see Figure 4 The fourth embodiment of the dialogic career interest dynamic correction method in this invention includes: 401. Calculate the interest matching degree between the interest information in the converged profile and the job nodes in the job knowledge graph; In this embodiment, the system uses cosine similarity to calculate the matching degree between the interest vector (such as the six-type score vector) in the converged profile and the interest vector of the job node in the job knowledge graph. The matching degree score is normalized to the range of [0,1], and the higher the score, the more compatible the interests are.

[0049] 402. Calculate the skill matching degree between the ability information in the convergent profile and the job nodes in the job knowledge graph; In this embodiment, the system calculates the matching degree between the skill set in the converged profile and the skill requirements of the job node based on Jaccard similarity or cosine similarity. For partially matched skills, a weighted calculation is used (such as core skills having a higher weight).

[0050] 403、evaluate the compatibility of the constraint information in the converged image with the constraint of the post node in the post knowledge graph, the constraint compatibility including geographical location coincidence degree and salary expectation coincidence degree; In this embodiment, the system evaluates the constraint compatibility, the geographical location coincidence degree is based on distance calculation (such as the distance between the post location and the preferred city, the smaller the distance, the higher the score), and the salary expectation coincidence degree is based on salary range overlap (such as whether the post salary meets the expected lower limit), and the compatibility score is calculated by a rule engine.

[0051] 404、based on the preset weight proportion, the interest matching degree, the skill matching degree and the constraint compatibility are weighted and fused to calculate a comprehensive recommendation score; In this embodiment, the system uses a preset weight proportion (such as interest 0.3, skill 0.5, and constraint 0.2) to weighted sum the three matching degrees to obtain a comprehensive recommendation score, and the weight can be dynamically adjusted according to user feedback, and the score is used for post sorting.

[0052] 405、according to the comprehensive recommendation score, the multiple candidate posts are sorted to obtain candidate post sorting information, and a post recommendation list is generated as post recommendation information according to the candidate post sorting information; In this embodiment, the system sorts the candidate posts in descending order according to the comprehensive recommendation score to generate candidate post sorting information, and the post recommendation list includes post name, company, matching score and key label, and is output in JSON or table format.

[0053] 406、based on the post recommendation information, an interpretable recommendation reason is generated, the recommendation reason including interest matching explanation, skill gap analysis and constraint satisfaction situation; In this embodiment, the system generates a recommendation reason in natural language format based on the post recommendation information, the interest matching explanation describes the interest matching point (such as “the post requires research characteristics, and your evaluation shows that you are good at analysis”), the skill gap analysis lists the missing skills (such as “you need to learn project management”), and the constraint satisfaction situation summarizes the location and salary coincidence degree, and the reason is generated using a template filling or an NLG model (such as GPT series).

[0054] 407、the post recommendation information and the recommendation reason are sent to the user end corresponding to the student, so that the user end generates and displays a visual recommendation page; In this embodiment, the system sends the post recommendation information and the recommendation reason to the user end (such as Web or App) through RESTful API, the user end parses the data using a front-end framework (such as React or Vue), generates a visual recommendation page, including post card, matching degree chart and reason text, and improves user experience.

[0055] 408、receive feedback information of the user terminal, the feedback information including recommendation satisfaction and the user's final job selection; In this embodiment, the system collects feedback information through the user terminal interface, including recommendation satisfaction (such as 1-5 score) and the user's final job selection (such as clicking on the application or ignoring), and the feedback data is stored for model optimization.

[0056] 409、adjust the weight proportion used for weighted fusion calculation of interest matching degree, skill matching degree and constraint compatibility according to the feedback information; In this embodiment, the system uses feedback information to adjust the weight proportion through reinforcement learning or A / B testing, for example, if the user is more satisfied with the interest matching, the interest weight is increased, otherwise the skill or constraint weight is adjusted to make the recommendation more personalized.

[0057] In the embodiment of the application, by calculating the interest matching degree, the skill matching degree and evaluating the constraint compatibility, an evaluation system covering the core dimensions of the person-job matching is constructed, and a comprehensive recommendation score is obtained based on the preset weight for weighted fusion, the precise sorting and recommendation list generation of the candidate post are realized, the transparency and persuasiveness of the recommendation result are greatly enhanced by generating an interpretable recommendation reason including interest matching explanation, skill gap analysis and constraint satisfaction, which helps students understand the recommendation logic and make more intelligent decisions, the intuitiveness of information acquisition and user experience are improved by sending the recommendation result and reason to the user terminal for visual display, finally, the weight proportion in the matching degree calculation is dynamically adjusted by collecting the satisfaction feedback and actual selection behavior of the user terminal, forming a complete recommendation effect evaluation and model optimization closed loop, so that the system has the advantages of continuous self-evolution and continuously enhanced personalized recommendation ability.

[0058] The dialog type occupation interest dynamic correction method in the embodiment of the application is described above, and the dialog type occupation interest dynamic correction device in the embodiment of the application is described below, please refer to Figure 5 The dialog type occupation interest dynamic correction device in the embodiment of the application includes one embodiment: The collection guiding module 501 is used for collecting the preliminary occupation willingness information of the students, guiding the students to complete the Holland occupation interest evaluation based on the dialog type Agent, and obtaining the occupation interest evaluation result; The construction comparison module 502 is used for constructing a multi-dimensional user portrait according to the occupation interest evaluation result, comparing the occupation interest evaluation result with the preliminary occupation willingness information, and obtaining a comparison result; The difference identification and interaction generation module 503 is used for identifying the difference in the comparison result and generating a difference label, based on the difference label, through the multi-round interaction between the dialog type Agent and the students to explain the source of the difference and guide the students to clarify the real willingness, and obtaining multi-round interaction information; The correction module 504 is configured to dynamically correct the multi-dimensional user portrait according to the multi-round interaction information to obtain a convergent portrait, and construct a post knowledge graph based on the convergent portrait. The generation and output module 505 is configured to generate post recommendation information based on the post knowledge graph, in combination with interest matching degree, skill matching degree and constraint compatibility, and output an interpretable recommendation reason.

[0059] In the embodiment, the preliminary career aspiration information of the student is collected, the student is guided to complete the Holland career interest assessment based on the dialog Agent, the multi-dimensional user portrait is constructed according to the career interest assessment result, the career interest assessment result is compared with the preliminary career aspiration information, the difference in the comparison result is identified and a difference label is generated, the multi-round interaction between the student and the dialog Agent is performed based on the difference label, the multi-dimensional user portrait is dynamically corrected according to the multi-round interaction information, the authenticity of the user portrait is significantly improved, the error recommendation caused by self-reporting bias is prevented, the accuracy of post recommendation is improved, the post knowledge graph is constructed based on the convergent portrait, the post recommendation information is generated based on the post knowledge graph, in combination with interest matching degree, skill matching degree and constraint compatibility, and the interpretable recommendation reason is output, the transparency of the system is enhanced, and the user satisfaction is improved.

[0060] Please refer to Figure 6 Another embodiment of the dialog career interest dynamic correction device in the embodiment of the application includes: The collection and guidance module 501 is configured to collect the preliminary career aspiration information of the student, guide the student to complete the Holland career interest assessment based on the dialog Agent, and obtain a career interest assessment result. The construction and comparison module 502 is configured to construct a multi-dimensional user portrait according to the career interest assessment result, compare the career interest assessment result with the preliminary career aspiration information, and obtain a comparison result. The identification and generation interaction module 503 is configured to identify the difference in the comparison result and generate a difference label, perform multi-round interaction with the student based on the difference label through the dialog Agent to explain the source of the difference and guide the student to clarify the real aspiration, and obtain multi-round interaction information. The correction module 504 is configured to dynamically correct the multi-dimensional user portrait according to the multi-round interaction information to obtain a convergent portrait, and construct a post knowledge graph based on the convergent portrait. The generation and output module 505 is configured to generate post recommendation information based on the post knowledge graph, in combination with interest matching degree, skill matching degree and constraint compatibility, and output an interpretable recommendation reason. In the embodiment, the collection guidance module 501 comprises: a collection unit 5011, configured to collect preliminary career aspiration information of a student through a natural language interaction interface, the preliminary career aspiration information comprising career preference, skill self-evaluation and career values; a generation guidance unit 5012, configured to generate interactive questions based on the Holland Career Interest Inventory Scale using a conversational Agent, and guide the student to complete the interactive questions through multiple rounds of dialogue to obtain answer information of the student; and an analysis and calculation unit 5013, configured to analyze the answer information using a natural language processing technology and calculate Holland six-type interest scores to obtain a career interest evaluation result.

[0061] In the embodiment, the comparison construction module 502 comprises: an extraction and construction unit 5021, configured to extract interest dimension features based on the career interest evaluation result, and construct a multi-dimensional user portrait according to the interest dimension features and personal basic information of the student; an alignment calculation unit 5022, configured to perform semantic alignment on the career interest evaluation result and the preliminary career aspiration information to obtain an alignment result, and calculate an interest consistency score according to the alignment result; and a first generation unit 5023, configured to generate a comparison result when the interest consistency score is lower than a preset score threshold, the comparison result comprising consistent items, inconsistent items and a difference degree.

[0062] In the embodiment, the recognition and generation interaction module 503 comprises: a recognition unit 5031, configured to identify a difference type in the comparison result using a rule engine and a machine learning model, the difference type being one or more of interest preference conflict, skill cognitive bias and insufficient career information; a generation and construction unit 5032, configured to generate a difference label for each difference type and construct a difference explanation template; an interaction unit 5033, configured to perform multiple rounds of interaction with the student through the conversational Agent based on the difference label and the difference explanation template to provide a difference source analysis and guide the student to clarify the real aspiration; and a recording and extraction unit 5034, configured to record dialogue content of the multiple rounds of interaction, and extract key clarification information in the dialogue content as multi-round interaction information.

[0063] In the embodiment, the revision construction module 504 comprises: an updating unit 5041, configured to update interest weights and skill labels in the multi-dimensional user portrait according to the multi-round interaction information using an incremental learning algorithm to obtain a converged portrait; an extraction unit 5042, configured to extract post entities, skill requirements and interest matching conditions from public post descriptions of a career database; a construction unit 5043, configured to construct a post knowledge graph according to the post entities, the skill requirements and the interest matching conditions, the post knowledge graph comprising post nodes, skill nodes, interest nodes and relationship edges; a writing unit 5044, configured to write interest information, ability information and constraint information in the converged portrait into the post knowledge graph; and a conversion unit 5045, configured to convert the post knowledge graph into a semantic vector using a graph embedding technology.

[0064] In the embodiment, the generating output module 505 includes: a first calculation unit 5051 configured to calculate an interest matching degree of the interest information in the converged portrait and a post node in the post knowledge graph; a second calculation unit 5052 configured to calculate a skill matching degree of the ability information in the converged portrait and the post node in the post knowledge graph; an evaluation unit 5053 configured to evaluate a constraint compatibility of the constraint information in the converged portrait and the post node in the post knowledge graph, the constraint compatibility including a geographical position coincidence degree and a salary expectation coincidence degree; a third calculation unit 5054 configured to perform weighted fusion calculation on the interest matching degree, the skill matching degree, and the constraint compatibility based on a preset weight proportion to obtain a comprehensive recommendation score; a sorting generating unit 5055 configured to sort the plurality of candidate posts according to the comprehensive recommendation score to obtain candidate post sorting information, and generate a post recommendation list as the post recommendation information according to the candidate post sorting information; and a second generating unit 5056 configured to generate an interpretable recommendation reason based on the post recommendation information, the recommendation reason including an interest matching explanation, a skill gap analysis, and a constraint satisfaction situation.

[0065] In the embodiment, the method further includes: a sending module 506 configured to send the post recommendation information and the recommendation reason to a user terminal corresponding to the student, so that the user terminal generates and displays a visual recommendation page; a receiving module 507 configured to receive feedback information of the user terminal, the feedback information including a recommendation satisfaction degree and a final post selection of the user; and an adjusting module 508 configured to adjust a weight proportion used for weighted fusion calculation on the interest matching degree, the skill matching degree, and the constraint compatibility according to the feedback information.

[0066] The above Figure 5 and Figure 6 The dialogical professional interest dynamic correction device in the embodiment is described in detail from the perspective of a modular functional entity, and the dialogical professional interest dynamic correction device in the embodiment is described in detail from the perspective of hardware processing.

[0067] Figure 7is a structural schematic view of a dialog-based dynamic occupation interest correction device provided by an embodiment of the present application. The dialog-based dynamic occupation interest correction device 600 can have great differences due to different configurations or performances, and can include one or more central processing units (CPUs) 610 (for example, one or more processors) and a memory 620, one or more storage media 630 (for example, one or more mass storage devices) storing an application program 633 or data 632. The memory 620 and the storage media 630 can be temporary storage or persistent storage. The program stored in the storage media 630 can include one or more modules (not shown in the figure), and each module can include a series of instruction operations in the dialog-based dynamic occupation interest correction device 600. Further, the processor 610 can be configured to communicate with the storage media 630 and execute the series of instruction operations in the storage media 630 on the dialog-based dynamic occupation interest correction device 600 to implement the steps of the dialog-based dynamic occupation interest correction method provided by the above-mentioned method embodiments.

[0068] The dialog-based dynamic occupation interest correction device 600 can also include one or more power supplies 640, one or more wired or wireless network interfaces 650, one or more input / output interfaces 660, and / or one or more operating systems 631, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, and the like. Those skilled in the art can understand that the dialog-based dynamic occupation interest correction device 600 can also include other components, and the components shown in the figure are not intended to limit the dialog-based dynamic occupation interest correction device. Figure 7 The dialog-based dynamic occupation interest correction device structure shown does not constitute a limitation based on the dialog-based dynamic occupation interest correction device, and can include more or fewer components than shown, or combine certain components, or different component arrangements.

[0069] The present application also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions, and when the instructions are run on a computer, the computer executes the steps of the dialog-based dynamic occupation interest correction method.

[0070] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system or device, unit can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0071] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or say the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0072] Finally, it should be noted that: the above only for the preferred examples of the present application, and not for limiting the present application, although the present application is described in detail with reference to the foregoing embodiments, for those skilled in the art, it still can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A dynamic dialogue method for correcting career interest, characterized in that, The method comprises the following steps: Collecting the initial career aspiration information of students, guiding the students to complete the Holland career interest assessment based on a conversational agent, and obtaining the career interest assessment results; Constructing a multi-dimensional user portrait based on the career interest assessment results, comparing the career interest assessment results with the initial career aspiration information, and obtaining the comparison results; Identifying the differences in the comparison results and generating difference labels, based on the difference labels, conducting multi-round interactions with the students through a conversational agent to explain the sources of differences and guide the students to clarify the real intentions, and obtaining multi-round interaction information; According to the multi-round interaction information, dynamically correcting the multi-dimensional user portrait to obtain a convergent portrait, and constructing a post knowledge graph based on the convergent portrait; Based on the post knowledge graph, combining interest matching degree, skill matching degree and constraint compatibility to generate post recommendation information, and outputting an interpretable recommendation reason.

2. The dynamic revision method of conversational career interest according to claim 1, characterized in that, The method comprises the following steps: Collecting the initial career aspiration information of students, guiding the students to complete the Holland career interest assessment based on a conversational agent, and obtaining the career interest assessment results; Collecting the initial career aspiration information of students through a natural language interaction interface, the initial career aspiration information including career preference, skill self-evaluation and career values; Using a conversational agent to generate interactive questions based on the Holland career interest assessment scale, and guiding the students to complete the interactive questions through multi-round dialogue to obtain the students' answer information; 3. The method of claim 1, wherein, Using natural language processing technology to analyze the answer information and calculate the Holland six-type interest score to obtain the career interest assessment results. The method comprises the following steps: Based on the career interest assessment results, extracting interest dimension features, and constructing a multi-dimensional user portrait based on the interest dimension features and the students' personal basic information; Aligning the career interest assessment results with the initial career aspiration information semantically to obtain an alignment result, and calculating an interest consistency score based on the alignment result; 4. The dialogic career interest dynamic correction method according to claim 1, characterized in that, When the interest consistency score is lower than a preset score threshold, a comparison result is generated, which includes consistent items, inconsistent items and difference degrees. The method comprises the following steps: Using a rule engine and a machine learning model to identify the difference types in the comparison results, the difference types being one or more of interest preference conflict, skill cognitive bias and insufficient career information; Generating a difference label for each difference type and constructing a difference explanation template; Conducting multi-round interactions with the students through a conversational agent based on the difference label and the difference explanation template to provide difference source analysis and guide the students to clarify the real intentions; Recording the dialogue content of multi-round interactions, and extracting the key clarification information in the dialogue content as multi-round interaction information.

5. The dialogic career interest dynamic correction method according to claim 1, characterized in that, The multi-dimensional user portrait is dynamically corrected according to the multi-round interaction information to obtain a convergent portrait, and a post knowledge graph is constructed based on the convergent portrait, including: An interest weight and a skill label in the multi-dimensional user portrait are updated using an incremental learning algorithm according to the multi-round interaction information to obtain a convergent portrait; Post entities, skill requirements and interest matching conditions are extracted from public post descriptions in a career database; A post knowledge graph is constructed according to the post entities, the skill requirements and the interest matching conditions, and the post knowledge graph includes post nodes, skill nodes, interest nodes and relationship edges; Interest information, ability information and constraint information in the convergent portrait are written into the post knowledge graph; The post knowledge graph is converted into a semantic vector using a graph embedding technique.

6. The method of claim 1, wherein, Based on the post knowledge graph, post recommendation information is generated in combination with an interest matching degree, a skill matching degree and constraint compatibility, and an interpretable recommendation reason is output, including: An interest matching degree of interest information in the convergent portrait and a post node in the post knowledge graph is calculated; A skill matching degree of ability information in the convergent portrait and a post node in the post knowledge graph is calculated; Constraint compatibility of constraint information in the convergent portrait and a post node in the post knowledge graph is evaluated, and the constraint compatibility includes a geographical location compliance degree and a salary expectation compliance degree; The interest matching degree, the skill matching degree and the constraint compatibility are weighted and fused based on a preset weight proportion to obtain a comprehensive recommendation score; A plurality of candidate posts are sorted according to the comprehensive recommendation score to obtain candidate post sorting information, and a post recommendation list is generated as post recommendation information according to the candidate post sorting information; An interpretable recommendation reason is generated based on the post recommendation information, and the recommendation reason includes interest matching explanation, skill gap analysis and constraint satisfaction situation.

7. The method of claim 1, wherein the dynamic revision of the career interest conversation is based on a user's current life situation. After the post knowledge graph is constructed, the interest matching degree, the skill matching degree and the constraint compatibility are combined to generate post recommendation information, and an interpretable recommendation reason is output, and the method further includes: The post recommendation information and the recommendation reason are sent to a user terminal corresponding to the student, so that the user terminal generates and displays a visual recommendation page; Feedback information of the user terminal is received, and the feedback information includes recommendation satisfaction and a user's final post selection; The weight proportion used for weighted fusion calculation of the interest matching degree, the skill matching degree and the constraint compatibility is adjusted according to the feedback information.

8. A dialogue device for dynamically revising a career interest, characterized by, It includes: A collection guiding module is used to collect preliminary career aspiration information of a student, guide the student to complete a Holland career interest evaluation based on a dialog Agent, and obtain a career interest evaluation result; A construction comparison module is used to construct a multi-dimensional user portrait according to the career interest evaluation result, compare the career interest evaluation result with the preliminary career aspiration information, and obtain a comparison result; The recognition generation interaction module is configured to recognize the difference in the comparison result and generate a difference label, based on the difference label, perform multi-round interaction with the student through a dialog Agent to explain the source of the difference and guide the student to clarify the real intention, and obtain multi-round interaction information; The correction modeling module is configured to dynamically correct the multi-dimensional user portrait according to the multi-round interaction information, obtain a convergent portrait, and construct a post knowledge graph based on the convergent portrait; The generation output module is configured to generate post recommendation information based on the post knowledge graph, in combination with interest matching degree, skill matching degree, and constraint compatibility, and output an interpretable recommendation reason.

9. A dialogue device for dynamically revising a career interest, characterized by, The dialog-based professional interest dynamic correction device includes a memory and at least one processor, and the memory stores instructions; The at least one processor invokes the instructions in the memory, so that the dialog-based professional interest dynamic correction device performs each step of the dialog-based professional interest dynamic correction method according to any one of claims 1-7.

10. A computer-readable storage medium having stored thereon instructions, the computer-readable storage medium comprising: The instructions are executed by the processor to implement each step of the dialog-based professional interest dynamic correction method according to any one of claims 1-7.

Images