An enterprise recruitment AI intelligent recommendation method and system based on behavior portrait

Abstract

The application relates to an enterprise recruitment AI intelligent recommendation method and system based on a behavior portrait, which comprises the following steps: obtaining resume text information and post demand information, generating an interactive task and obtaining tendency interaction information; performing time sequence slicing on the tendency interaction information, setting an exponential decay weight and a dynamic enhancement weight, and generating enhanced interaction information; extracting a target feature set, constructing a personal comprehensive portrait of a target object, and constructing a post demand portrait of a target enterprise; inputting the personal comprehensive portrait and the post demand portrait into an analysis model to generate an adaptability score; generating recommendation information based on the adaptability score as matching result information and sending the matching result information to the target enterprise and the target object respectively; and in summary, the application has the effects of improving the accuracy and stability of the man-job matching by generating interactive tasks to obtain tendency interaction information, performing time sequence slicing and weighting, extracting a target feature set, constructing a portrait, and generating an adaptability score.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and in particular to an AI-powered intelligent recommendation method and system for enterprise recruitment based on behavioral profiling. Background Technology

[0002] In the current corporate recruitment and talent recommendation process, AI-driven intelligent matching systems are widely used in the process of matching job seekers with positions.

[0003] Mainstream technical solutions primarily rely on keyword parsing and skill tag matching of job seekers' resumes, or on collaborative filtering algorithms based on historical application and hiring data for recommendations. However, these existing methods suffer from the following technical shortcomings: First, the data sources relied upon by existing technologies are typically static and isolated, mainly limited to one-time resume texts. They lack the acquisition and fusion of temporal interaction behavior sequences that can dynamically reflect user intentions, resulting in weak representational capabilities of the extracted feature vectors. This makes it difficult to accurately capture the evolution of job seekers' preferences and real-time changes in decision-making styles, leading to one-sided input information and limited generalization ability of the analysis model. Second, existing technologies lack differentiation of job seekers' historical interaction behaviors and fail to establish a dynamic evaluation of the timeliness of behavioral data and critical decision-making periods. The assessment mechanism is prone to misjudging the current status and long-term development potential of job seekers, affecting the timeliness and foresight of the matching. In addition, traditional analysis models are mostly shallow models based on keyword matching or collaborative filtering models. They usually focus too much on the surface connection of explicit skills and experience, and generally lack the ability to explore the implicit characteristics of job seekers in simulated work situations, such as decision-making paths, coping patterns and value expressions. This makes the recommendation results often stay at the level of skill matching, and it is difficult to establish a non-linear relationship between ability, inclination and deep job requirements. As a result, it is impossible to deeply assess the fit between the person and the job in terms of soft skills, team culture adaptability and long-term career development potential. This leads to the matching assessment remaining at the level of surface skill comparison, and the output analysis and assessment results have low accuracy. Summary of the Invention

[0004] To address the aforementioned shortcomings, this application provides an AI-powered intelligent recommendation method and system for enterprise recruitment based on behavioral profiles.

[0005] The above-mentioned objective of this application is achieved through the following technical solution:

[0006] A method for AI-powered intelligent recommendation in enterprise recruitment based on behavioral profiles includes the following steps:

[0007] Obtain the resume text information of the target object, and obtain the job requirements information of the target company based on the resume text information;

[0008] An interactive task is generated based on the resume text information and job requirements information and provided to the target object. The target object's tendency interaction information based on the interactive task is collected. The tendency interaction information includes feedback content and process data collected through the interactive task.

[0009] The interactive information is sliced ​​according to the timestamp, and the interactive behavior in the first preset time period is given an exponential decay weight, and the interactive behavior in the second preset time period is given a dynamic enhancement weight to generate enhanced interactive information.

[0010] A target feature set is extracted based on enhanced interactive information and resume text information. The target feature set includes tendency features, ability features, and implicit decision features.

[0011] A comprehensive personal profile of the target object is constructed based on the target feature set, and a job requirement profile of the target company is constructed based on the job requirement information.

[0012] Input the individual's comprehensive profile and the job requirement profile into the pre-trained analysis model to generate an adaptive score;

[0013] Recommendations are generated based on the adaptability score, including key adaptability points. These recommendations are then sent as matching results to the target company and the target individual.

[0014] The second objective of this invention is achieved through the following technical solution:

[0015] An AI-powered intelligent recommendation system for enterprise recruitment based on behavioral profiles includes:

[0016] The information acquisition module is used to acquire the resume text information of the target object, and based on the resume text information, to acquire the job requirements information of the target company.

[0017] The task generation module is used to generate interactive tasks based on resume text information and job requirement information and provide them to the target object, and to collect the target object's tendency interaction information based on the interactive tasks. The tendency interaction information includes feedback content and process data collected through the interactive tasks.

[0018] The information enhancement module is used to perform time-series slicing of the interactive information according to timestamps, set exponential decay weights for interactive behaviors in a preset first time period, set dynamic enhancement weights for interactive behaviors in a preset second time period, and generate enhanced interactive information.

[0019] The feature extraction module is used to extract a target feature set based on enhanced interaction information and resume text information. The target feature set includes tendency features, ability features, and implicit decision features.

[0020] The profile building module is used to build a comprehensive personal profile of the target object based on the target feature set, and to build a job requirement profile of the target company based on job requirement information;

[0021] The scoring generation module is used to input the individual's comprehensive profile and the job requirement profile into the pre-trained analysis model to generate an adaptive score;

[0022] The information generation module is used to generate recommendation information including adaptation points based on the adaptability score, and send the recommendation information as matching result information to the target enterprise and the target object respectively.

[0023] This application also relates to a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described AI-powered intelligent recommendation method for enterprise recruitment based on behavioral profiles.

[0024] This application also relates to a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described AI-powered intelligent recommendation method for enterprise recruitment based on behavioral profiles.

[0025] In summary, the AI-powered intelligent recommendation method and system for enterprise recruitment based on behavioral profiling provided in this application can capture dynamic behavioral changes of job seekers and mine implicit decision-making features by generating interactive tasks to obtain tendency interaction information, performing time-series slice weighting, extracting target feature sets and constructing profiles, and generating adaptive scores. This can solve the technical problem in existing technologies that cannot effectively assess the real-time tendency and suitability of job seekers, and improve the accuracy and stability of job matching. Attached Figure Description

[0026] Figure 1 This is a flowchart of an embodiment of an AI-powered intelligent recommendation method for enterprise recruitment based on behavioral profiling, as described in this application.

[0027] Figure 2 This is a flowchart of step S10 in an embodiment of an AI-powered intelligent recommendation method for enterprise recruitment based on behavioral profiles, as described in this application.

[0028] Figure 3 This is a flowchart of step S20 in an embodiment of an AI-powered intelligent recommendation method for enterprise recruitment based on behavioral profiling, as described in this application. Detailed Implementation

[0029] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0030] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0031] In one embodiment, this application discloses an AI-powered intelligent recommendation method for enterprise recruitment based on behavioral profiles, such as... Figure 1 As shown, the specific steps include the following:

[0032] S10: Obtain the resume text information of the target object, and obtain the job requirements information of the target company based on the resume text information;

[0033] In this embodiment, resume text information refers to structured or unstructured text data provided by the target applicant to describe their educational background, work experience, skills and expertise, project experience, and job intentions; job requirement information refers to information set by the company for different positions to describe the job responsibilities, qualifications, skill requirements, experience requirements, and other soft requirements. The provision of resume text information has been authorized by the target applicant.

[0034] S20: Generate an interactive task based on resume text information and job requirement information and provide it to the target object, and collect the target object's tendency interaction information based on the interactive task. The tendency interaction information includes feedback content and process data collected through the interactive task.

[0035] In this embodiment, interactive tasks refer to a series of simulated scenarios or questions dynamically generated based on the target object's resume information and job requirements information. These tasks aim to guide the target object to perform interactive operations or provide feedback. The purpose of interactive tasks is to deeply explore the target object's behavioral tendencies, decision-making patterns, and implicit characteristics. Tendency interaction information refers to the behavioral data generated when the target object interacts with the interactive task. This data can reflect the target object's interests, behavioral patterns, and potential career inclinations to a certain extent. For example, if the interactive task is an online test, the user's feedback and operational data can be considered as tendency interaction information. Furthermore, online assessment interactive tasks including scenario simulation, value ranking, and virtual project collaboration can be provided to the target object. All the target object's answers, operation order, hesitation time, and modification records in this interactive task constitute tendency interaction information. The acquisition of this tendency interaction information requires prior authorization from the target object and relevant parties. Furthermore, this interactive task can be designed based on a job competency model, a cognitive behavior analysis framework, and actual needs, and stable behavioral signals can be collected through a standardized interaction process.

[0036] It should be noted that the interactive task in this embodiment is not limited to a single, short-term interactive task, but can be further regarded as a multi-stage career exploration interactive task that lasts for a period of time. For example, the multi-stage career exploration interactive task may include: periodic tendency review task: inviting the target object to re-rank values ​​or select scenarios every day, three days or a week, and informing the target object that they can make corrections at any time before the deadline to track the stability and trend of their tendency; targeted information feedback and re-interaction: according to the target object's initial profile, regularly push relevant industry articles, job interpretations and corresponding tests, and record their browsing time and answering behavior, etc. At the same time, the tendency interaction information is entirely based on the data generated by the user's active interaction in the multi-stage career exploration interactive task, and does not involve the capture of behavior logs from any third-party platform.

[0037] S30: Slice the tendency interaction information into time sequence according to the timestamp, set the exponential decay weight for the interaction behavior in the preset first time period, set the dynamic enhancement weight for the interaction behavior in the preset second time period, and generate enhanced interaction information.

[0038] In this embodiment, time-series slicing refers to dividing continuous, chronologically arranged interactive behavior data into several discrete time segments according to a preset time interval or strategy. This aims to differentiate the behavioral data within different time periods to capture the timeliness and evolution trend of the behavior. Exponential decay weighting assigns lower weights to earlier interactive behaviors, and this exponential decay weight decreases exponentially with the increase of time intervals, aiming to reflect the timeliness of the behavior, i.e., recent behaviors have a greater impact on the current state. Dynamic enhancement weighting assigns higher weights to interactive behaviors within a specific time period, such as reflecting stages of deep thinking or key decisions. Dynamic enhancement weighting can be dynamically adjusted according to factors such as the density, modification frequency, or correlation strength of interactive behaviors to represent the value of important behaviors. Enhanced interactive information refers to weighted interactive data that more accurately reflects the current state and deep intentions of the target object after time-series slicing of the tendency interactive information and applying exponential decay weighting and dynamic enhancement weighting.

[0039] S40: Extract a target feature set based on the enhanced interactive information and resume text information. The target feature set includes tendency features, ability features, and implicit decision features.

[0040] In this embodiment, the target feature set refers to the set of key attributes extracted from enhanced interactive information and resume text information to describe the target object. The target feature set includes tendency features, ability features, and implicit decision features. Tendency features refer to features that reflect the target object's career preferences, interests, values, and response patterns to specific work situations. Ability features refer to features that reflect the target object's knowledge, skills, and experience. Ability features are mainly extracted from resume text information and supplemented by performance in interactive tasks. Implicit decision features refer to features that reflect the target object's thinking style, decision-making style, risk preference, and problem-solving path when facing problems.

[0041] S50: Construct a comprehensive personal profile of the target object based on the target feature set, and construct a job requirement profile of the target company based on job requirement information;

[0042] In this embodiment, the personal comprehensive profile refers to a multi-dimensional digital description of the target object formed by integrating and vectorizing the tendency characteristics, ability characteristics, and implicit decision-making characteristics, aiming to accurately depict the comprehensive quality and potential value of the target object; the job requirement profile refers to a multi-dimensional digital description of the target job formed by encoding and combining job requirement information, aiming to accurately depict the real requirements of the job and the inherent characteristics of the enterprise.

[0043] S60: Input the individual's comprehensive profile and the job requirement profile into the pre-trained analysis model to generate an adaptive score;

[0044] In this embodiment, the analysis model refers to a pre-trained machine learning model that receives a comprehensive personal profile and a job requirement profile as input and outputs an adaptability score. This analysis model is typically built based on deep learning or complex statistical methods and is used to evaluate the person-job fit. The adaptability score is a numerical value output by the analysis model that quantifies the degree of matching between the target object and the target job. A higher adaptability score indicates a higher degree of matching.

[0045] S70: Generate recommendation information including adaptation points based on the adaptability score, and send the recommendation information as matching result information to the target enterprise and the target object respectively.

[0046] In this embodiment, recommendation information refers to text or structured data generated based on the adaptability score and containing adaptability points, which aims to provide target enterprises and target objects with detailed descriptions and suggestions about the matching results; matching result information refers to the matching feedback results sent to target enterprises and target objects.

[0047] Specifically, the process involves obtaining the resume text information of the target individual and then using this resume text information to obtain the corresponding job requirements information of the target company. In one implementation, the resume text information can be obtained by the target individual manually uploading their resume file and then parsing the resume file. The job requirements information can be provided by the target company. In another implementation, the job requirements information can be obtained from the company's official website or recruitment website and then processed in a structured manner.

[0048] Furthermore, based on resume text information and job requirements, interactive tasks are generated and provided to the target audience, and the target audience's tendency to interact based on the interactive tasks is collected. One way to generate interactive tasks is to randomly extract several general questions or situational questions from a preset task library according to keywords in the resume and requirements in the job description, and combine them into a task sequence. When the target audience completes the task sequence, their selections, inputs, and other feedback, as well as process data such as operation time and click order, are recorded. For example, a multiple-choice question can be generated to ask the target audience how to handle a specific work conflict situation, and the selected options and the time spent completing the multiple-choice question can be recorded.

[0049] Furthermore, the interactive information is slicing according to timestamps, with exponential decay weights assigned to interactive behaviors within a preset first time period and dynamic enhancement weights assigned to interactive behaviors within a preset second time period, thereby generating enhanced interactive information. One method of time slicing is to arrange the timestamps of all interactive behaviors sequentially and divide them at fixed time intervals. For interactive behaviors within the first time period that are far from the current time, a fixed decay coefficient can be applied for weighting. For interactive behaviors within the second time period that are closer to the current time, a fixed enhancement weight can be assigned based on whether it is the last modification or whether the dwell time exceeds a certain threshold. For example, if the target object makes multiple modifications to a certain option in the task, only the behavior before the last modification can be given a lower weight, while the behavior after the last modification can be given a higher weight.

[0050] Furthermore, based on the enhanced interactive information and resume text information, a target feature set is extracted. One feature extraction method is to identify skill vocabulary and educational background from the resume text as ability features through keyword matching; to identify tendency features from the feedback content in the enhanced interactive information by matching the text with a pre-defined tendency dictionary; and to identify implicit decision-making features from the process data in the enhanced interactive information by statistically analyzing the number of operation steps and the average dwell time. For example, if the resume contains words such as "Python programming" and "data analysis," they are considered ability features; if feedback on interactive tasks includes words such as "teamwork" or "innovative thinking," they are considered tendency features; and if the target person spends a long time and makes multiple attempts on a complex task, they may be considered to have the implicit decision-making feature of "prudent decision-making."

[0051] Furthermore, a comprehensive personal profile of the target object is constructed based on the target feature set, and a job requirement profile of the target company is constructed based on the job requirement information. One way to construct the profile is to concatenate all extracted tendency features, ability features, and implicit decision features into a long vector as the comprehensive personal profile. For the job requirement profile, all text information in the job description can be processed using a bag-of-words model to generate a job feature vector. For example, the comprehensive personal profile can be a list containing tags such as "Python skills", "data analysis ability", "teamwork tendency", and "prudent decision-making style", while the job requirement profile can be a list containing tags such as "requires Python", "requires data analysis", and "requires teamwork".

[0052] Furthermore, the comprehensive personal profile and the job requirement profile are input into a pre-trained analysis model to generate an adaptability score. One analysis model can be a rule-based matching system. For example, if the skills in the personal profile have an 80% overlap with the requirements in the job profile, a higher adaptability score is given. This analysis model can pre-set several matching rules based on expert experience. For example, if the personal profile contains all the "hard skills" listed in the job profile, the initial score will be higher.

[0053] Finally, based on the adaptability score, recommended information including adaptability points is generated, and the recommended information is sent as matching results to the target company and the target object respectively. One way to generate recommended information is to output text prompts of "high matching degree" or "low matching degree" according to the adaptability score. For adaptability points, the skill points with the highest matching degree between the personal profile and the job profile can be listed. For example, if the adaptability score is 85 points, the recommended information "high matching degree, the main advantages are Python skills and data analysis ability" can be generated.

[0054] For example, suppose a user is looking for a senior data analyst position, and a company is recruiting for that position. First, the system obtains the user's resume text information, which includes their educational background, past work experience, programming languages ​​(such as Python), data analysis tools (such as SQL, Tableau), and project experience. Based on this resume information, the system retrieves the company's requirements for the senior data analyst position from a pre-defined job information database. These requirements may include proficiency in Python and SQL, experience in big data processing, and good communication and collaboration skills.

[0055] Furthermore, interactive tasks are generated based on the user's resume text information and the job requirements for a senior data analyst. For example, several simulated scenario tasks may be generated and provided to the user at different time periods, requiring the user to analyze several sales reports containing abnormal data on the corresponding date and time and make improvement suggestions. The simulated scenario task may include multiple multiple-choice questions, fill-in-the-blank questions, and an open text input box. When the user completes these tasks, their selected answers, the text content they enter, as well as the timestamp of each click, the order of operations, the duration of the pause, and even the history of modifying the answers will be recorded, constituting the interactive information.

[0056] Furthermore, the collected tendency interaction information is processed. For example, when a user completes a sales report analysis task, they may quickly browse and make some preliminary selections in the first few minutes. These behaviors belong to the interaction behaviors within the preset first time period. Based on the interval between the timestamps of these behaviors and the current time, an exponential decay function is applied to calculate and set decay weights, so that earlier behaviors have less impact on the final evaluation. Furthermore, if a user completed the corresponding analysis task a week, five days, three days, and one day ago, respectively, and some answers changed significantly or the interaction behaviors during the process were abnormal, an exponential decay function can be applied to calculate and set decay weights, so that earlier tasks further away from the current time have lower weights. If a user spends a long time on a key analysis step and modifies their text input multiple times, these behaviors belong to the interaction behaviors within the preset second time period. Based on the density of these behaviors, the frequency of modification, and the strength of their relevance to the core issue, dynamic enhancement weights are calculated and set to represent the importance of deep thinking and key decision-making behaviors. Through this differentiated weighting process, enhanced interaction information that better reflects the user's true intentions and decision-making patterns is generated.

[0057] Furthermore, based on the enhanced interactive information and the user's resume text information, a target feature set is extracted. Specifically, explicit ability features such as "Python programming", "SQL query" and "big data processing" can be extracted from the resume text, and tendencies such as "strong logical analysis ability" and "emphasis on data-driven decision-making" can be extracted from the feedback content of the enhanced interactive information. Implicit decision-making features such as "prudent decision-making" and "strong problem-solving ability" can be extracted from the process data of the enhanced interactive information.

[0058] Furthermore, a comprehensive user profile is constructed based on the target feature set. For example, the correlation weight between the tendency feature of "strong logical analysis ability" and the ability feature of "SQL query" is calculated through a preset interaction attention mechanism. Based on this, these features are adaptively fused to form an integrated feature vector as the comprehensive user profile. At the same time, based on the job requirements information of "senior data analyst", features such as "proficient in Python", "proficient in SQL", "good communication and collaboration" and "adaptable to fast-paced work culture" are extracted and multi-layered vector encoding and combination are performed to construct the job requirements profile for this position.

[0059] Furthermore, the constructed comprehensive user profile and job requirement profile are input into a pre-trained analysis model. This analysis model can compare "Python programming" in the user profile with "proficient in Python" in the job profile, and compare "strong logical analysis ability" in the user profile with "good communication and collaboration" in the job profile. It also calculates the initial fit of the above comparisons and generates a quantitative adaptability score. For example, 88 points indicates that the user is highly matched with the job.

[0060] Finally, recommendations are generated based on the adaptability score. These recommendations include specific adaptability points, such as "This user's Python and SQL skills are highly matched to the job requirements, and they have demonstrated excellent logical analysis and prudent decision-making abilities through interactive tasks, which are highly consistent with the job's requirements for data-driven decision-making." These recommendations are then sent as matching results to both the user and the target company to help both parties understand the matching situation.

[0061] In existing technologies, the evaluation of users' job-seeking intentions is often limited to static resume text analysis, such as simply matching skill keywords in the resume with job requirements. However, this approach is difficult to capture the dynamic behavior and deep decision-making logic of users in real-world situations. This embodiment can overcome the limitations of traditional static data analysis by generating interactive tasks and collecting user feedback and process data in the tasks. For example, in the sales report analysis task mentioned above, traditional methods may only be able to identify the "Python" and "SQL" skills in the user's resume, but cannot know the user's specific analytical thinking and decision-making process when facing complex data.

[0062] Furthermore, existing technologies lack the ability to distinguish between the timeliness of behavioral data and critical decision-making periods, which may lead to biases in the judgment of the user's current state. This embodiment effectively solves this problem by performing time-series slicing of tendency interaction information and setting exponential decay weights and dynamic enhancement weights for interaction behaviors in different time periods. In the example, the behavior of users spending a long time and making multiple modifications on key analysis steps is given higher weight, which can more accurately capture the user's true intention of deep thinking and key decisions, rather than just superficial quick browsing behavior, so that the understanding of user behavior can be more in-depth and timely.

[0063] Furthermore, traditional analytical models primarily focus on the matching of explicit skills and experience, generally lacking the mining and analysis of job seekers' implicit characteristics. This embodiment, based on enhanced interactive information and resume text information, extracts tendency features, ability features, and implicit decision-making features to construct a multi-dimensional target feature set. In the example, in addition to extracting explicit ability features such as "Python programming" from the resume, it can also extract tendency and implicit decision-making features such as "strong logical analysis ability" and "prudent decision-making" from the interactive task. This allows the comprehensive personal profile to more fully reflect the user's soft skills, decision-making style, and value preferences, thereby enabling a deeper match with the job requirements profile. This mining and utilization of implicit features allows the final recommendation results to go beyond superficial skill matching, effectively assessing the fit between the person and the job in terms of soft skills and long-term development potential, thereby improving the overall processing efficiency of the talent recommendation process and the reliability of the adaptability score calculation results.

[0064] In one embodiment, step S10 includes:

[0065] S11: Obtain resume text information, which includes skills text information, experience text information, and intended job text information;

[0066] In this embodiment, resume text information refers to the comprehensive written expression of the target candidate's personal background, abilities, and job aspirations. It is divided into skills text information, experience text information, and desired job text information to structure the resume content, facilitating targeted analysis and extraction of key information. Skills text information typically includes the target candidate's professional skills and tool usage abilities; experience text information includes educational background, work experience, and project participation; and desired job text information directly reflects the target candidate's job search direction and expected job type. Step S11 specifically uses optical character recognition technology to convert paper or image-format resumes into editable text format, and then uses natural language processing technology to perform preliminary segmentation and classification of the text, identifying different types of information such as skills, experience, and desired job. Alternatively, a preset resume template or online form can be used to guide the target candidate to directly input structured information, ensuring that skills, experience, and desired job information are clearly distinguished upon submission.

[0067] S12: Extract keywords and analyze intent from the text information of intended job positions to generate a set of job intentions;

[0068] In this embodiment, the text information of the intended job posting directly reflects the job seeker's intention. Keyword extraction and intent parsing are performed to identify key elements such as job title, industry preference, job category, and salary expectation from the free text, and to understand the job seeker's intent corresponding to these elements, thereby forming a structured set of job intentions. Step S12 can adopt a rule-based keyword matching method, specifically by pre-setting an industry dictionary, a list of job titles, and skill tags, and extracting relevant keywords from the intended job posting text through regular expressions or pattern matching. At the same time, semantic analysis models are used to classify the intent of the text, such as determining whether it is "software development engineer" or "data analyst," and identifying their preference for the "Internet industry." Alternatively, machine learning models, such as support vector machines or recurrent neural networks, can be used to train the machine learning model with a large amount of labeled data, enabling it to identify key entities and parse job seeker intent from the intended job posting text.

[0069] Furthermore, the semantic analysis model or machine learning model that may be used in step S12 can be obtained and trained by those skilled in the art in combination with actual scenarios or needs.

[0070] S13: Based on the set of job preferences, perform matching and retrieval in the preset job information database to filter out the target company set and the corresponding job requirements information.

[0071] In this embodiment, a matching search is performed in the job information database based on the job intention set, aiming to find jobs that highly match the target's intention from a massive number of jobs, and further obtain the company information and detailed requirements of these jobs; step S13 can adopt a similarity matching method based on the vector space model, specifically, the keywords and intentions in the job intention set are vectorized, and each job description in the job information database is also vectorized, and then the cosine similarity or other distance measure between the vectors is calculated, and jobs with similarity higher than a preset similarity threshold are selected. The companies to which the selected jobs belong constitute a candidate company set, and their corresponding job requirement information is obtained; or, an inverted index or full-text search technology can be used to index all job descriptions in the job information database, and the keywords in the job intention set are used as query terms for retrieval. The retrieval results can be sorted according to the relevance score, and the most relevant jobs, their company information and job requirements can be selected.

[0072] Furthermore, the job information database can be stored in the form of triples (company ID-job ID-feature vector), where the job feature vector includes dimensions such as hard requirements, soft requirements, and implicit cultural characteristics. Each dimension is initially assigned a value according to its weight ratio, and it can be set accordingly based on the actual scenario and needs.

[0073] Specifically, the process involves acquiring the resume text information of the target candidate and structuring it into skills text information, experience text information, and intended job text information. Further, keyword extraction and intent analysis are performed on the intended job text information to identify key elements such as core job preferences and industry direction from the target candidate's expressed job-seeking intentions, and these are transformed into a structured set of job intentions. Finally, based on the generated set of job intentions, a matching search is performed in a pre-set job information database to accurately select candidate companies and their specific job requirements that highly match the target candidate's job-seeking intentions.

[0074] By using the above technical solutions, the resume text information is structured and segmented, and keywords are extracted and intent is analyzed from the text information of intended positions. This allows for the accurate capture of the target candidate's true job-seeking intentions and preferences. Furthermore, by matching and searching the job information database based on the generated set of job intentions, candidate companies and job requirements that highly match the target candidate's intentions can be selected. This provides highly relevant basic data for interactive task generation and personal profile construction, which can reduce the number of invalid recommendations and improve the accuracy of recommendations.

[0075] In one embodiment, step S20 includes:

[0076] S21: Analyze job requirements information to extract core competency requirements and work context elements;

[0077] In this embodiment, parsing job requirement information and extracting core competency requirements and work context elements refers to analyzing the job description text provided by the target company to identify the core competencies, professional skills requirements, and typical work environment or scenario characteristics necessary for the job. This can be achieved using methods such as text classification, information extraction, or rule matching. For example, by constructing a job competency ontology library, the text in the job description can be mapped to predefined core competencies and work context tags.

[0078] S22: Based on the resume text information, core competency requirements and work context elements, perform multi-dimensional matching in the preset task template library to select several related basic task templates;

[0079] In this embodiment, based on resume text information, core competency requirements, and work context elements, a multi-dimensional matching is performed in a pre-set task template library to select several related basic task templates. This means that various elements extracted from the resume and job requirements are used as matching conditions and compared with the basic task templates pre-stored in the task template library. Each task template in the task template library is associated with its assessable skills, experience, abilities, and contextual tags. The multi-dimensional matching can use a semantic similarity-based algorithm, such as calculating the vector distance between the extracted elements and the template tags, or a rule-based expert system to select the most relevant basic task templates based on the matching score.

[0080] Furthermore, the basic task template can be preset with three categories of tags: ability dimension, situation dimension, and difficulty dimension. The ability dimension corresponds to the core competency requirements of the position, the situation dimension corresponds to the elements of the work situation, and the difficulty dimension can be divided into beginner, intermediate, and advanced levels to match job seekers with different work experience. These can be set according to the actual scenario and needs.

[0081] S23: According to the preset evaluation logic rules, sort and logically connect the selected basic task templates to generate interactive tasks;

[0082] In this embodiment, the basic task templates selected are sorted and logically linked according to preset evaluation logic rules to generate interactive tasks. This means that after selecting multiple basic task templates, the selected basic task templates are organized and arranged according to preset evaluation logic rules. The evaluation logic rules may include making decisions based on the progression of task difficulty, the priority of skill assessment, the order of situation simulation, etc. For example, it can be set to conduct technical ability testing first, then situation simulation, and finally behavioral tendency assessment, thereby combining independent basic task templates to generate interactive tasks.

[0083] S24: Provide interactive tasks to the target object and record the target object's interaction log to form a tendency interaction information.

[0084] In this embodiment, providing interactive tasks to the target object and recording the target object's interaction log to constitute biased interaction information means presenting a pre-combined interactive task to the target object through a user interface and recording all the target object's operational behaviors during the task completion process. The interaction log is the core of the biased interaction information, including not only the target object's explicit feedback but also its implicit behavioral data. Feedback content includes the target object's selection of task options, sorting results, or text input content; this refers to explicit response data directly provided by the target object in the interactive task, such as the answers selected in a multiple-choice question, the order completed in a drag-and-drop sorting task, or in an open-ended question. The text responses entered in the question directly reflect the target audience's knowledge, perspectives, and decisions. Process data, including timestamps, operation sequences, dwell times, and modification history for each step, refers to the non-explicit behavioral data generated by the target audience during the completion of the interactive task. Timestamps record the precise time each operation occurred; operation sequences record the order in which the target audience performed the operations; dwell times record the time spent on a specific task or page; and modification history details the target audience's modifications to text input or option selections. This process data reveals the target audience's thinking patterns, decision-making hesitation, proficiency, and problem-solving strategies.

[0085] Specifically, the solution in this application analyzes job requirement information to identify the core competency requirements and work context elements of the target position. Based on resume information and the identified elements, it performs multi-dimensional matching in a pre-set task template library to select basic task templates that are highly relevant to the target object and job requirements. Subsequently, the basic task templates are sorted and logically linked through pre-set evaluation logic rules to generate highly targeted interactive tasks. The generated interactive tasks can effectively simulate the real work scenario of the target position, exploring the target object's professional skills, problem-solving abilities, behavioral tendencies, and decision-making patterns. During the execution of the interactive tasks, not only are the explicit feedback contents of the target object recorded, but also their implicit process data are recorded in detail, such as the timestamps of operation steps, operation sequences, dwell time, and modification history, so that the obtained tendency interaction information can more accurately reflect the target object's true abilities and potential traits.

[0086] For example, suppose a target applicant applies for a senior software engineer position. First, the job requirements for a "senior software engineer" position provided by the target company are analyzed to extract core competency requirements such as "proficient in Java concurrent programming," "capable of designing and optimizing high-concurrency systems," "familiar with distributed transaction processing," and "good teamwork and communication skills," as well as work context elements such as "rapid iteration," "high-pressure environment," and "cross-departmental collaboration." Simultaneously, the applicant's resume text information can be analyzed to extract key skill elements such as "Java," "SpringBoot," "microservice architecture," and "database optimization," as well as experience elements such as "responsible for backend development of a large e-commerce platform" and "leading the design of high-concurrency systems." Further, based on the extracted elements, multi-dimensional matching is performed in a pre-defined task template library. For example, it might filter out tasks such as "Java concurrent programming challenge," "microservice troubleshooting simulation," and "distributed transaction simulation." The system uses basic task templates such as "Scene Design" and "Team Collaboration and Communication Simulation" to generate complete interactive tasks. These templates are then sorted and logically linked according to pre-defined evaluation logic rules, such as "technical skills first, then soft skills, with increasing difficulty." For example, a "Java Concurrency Programming Challenge" might be presented first; if passed, the next step is "Microservice Troubleshooting Simulation," followed by "Team Collaboration and Communication Simulation." Once the interactive tasks are provided to the target audience, detailed interaction logs are recorded. For instance, in the "Java Concurrency Programming Challenge," the target audience's chosen solution, the code they wrote, the time spent writing the code, the number of code submissions, and the code modification process are all recorded. In the "Team Collaboration and Communication Simulation," the target audience's chosen communication strategy, their text responses, their thinking time, and the number of times they modified their responses are also recorded. This method allows for comprehensive and in-depth collection of the target audience's interactive tendencies.

[0087] Through the above technical solution, this application can generate interactive tasks closely related to the target object and job requirements, thereby enabling the target object to effectively demonstrate its true skills, abilities and behavioral tendencies; by recording explicit feedback content and implicit process data, this application can obtain more comprehensive tendency interaction information, thereby improving the accuracy of the assessment of the matching degree between the target object and the target job.

[0088] In one embodiment, step S30 includes:

[0089] S31: Extract the timestamps of all interactive behaviors in the tendency interaction information, form an interaction sequence, and divide the interaction sequence into several consecutive time segments based on the preset slice duration.

[0090] In this embodiment, step S31 aims to serialize and structure all behavioral events generated by the target object in the interactive task in terms of time. By extracting the time point of each interactive behavior, a complete behavioral timeline, i.e., an interaction sequence, can be constructed. The constructed continuous interaction sequence is divided into several discrete time segments according to a preset fixed or dynamic time length. This helps to independently analyze and weight the behavior in different time periods, thereby capturing the dynamic changes and temporal characteristics of the behavior. For example, the precise time of each click, input, selection, page jump, etc., can be recorded, and these timestamps can be associated with and stored with the corresponding behavioral events. When performing time-series slicing, a fixed slice duration can be set, for example, every 5 minutes or every 10 minutes, starting from the start time of the interaction sequence, to divide continuous time segments sequentially. Alternatively, when recording user behavior, a high-precision timestamp can be attached to each behavioral event. During processing, the slice duration can be adaptively determined according to the total duration of the interaction sequence and the expected analysis granularity to ensure that each time segment contains a certain number of behavioral samples.

[0091] S32: Determine the preset first time period and preset second time period in the time segment based on the preset time period division strategy;

[0092] In this embodiment, step S32 aims to identify and distinguish specific time periods with different processing logics within the divided time segments. The first preset time period typically refers to the period of relatively early, longer-interval interactive behaviors, whose importance may decrease over time. The second preset time period refers to a highly active period where recent or specific key interactive behaviors occur, and its importance may need to be increased. By dividing the time into the first and second preset time periods, different weights can be applied to behaviors within different time periods to reflect the value of the behaviors. Specific time period boundaries can be determined based on experience or data analysis. For example, the time period division can be dynamically determined based on the activity level of the interactive behavior or the task stage. Exploratory interactive behaviors at the beginning of a task can be classified as the first preset time period, while concentrated modification and confirmation behaviors before task submission can be classified as the second preset time period.

[0093] S33: For interactive behaviors within the preset first time period, based on the interval between their timestamp and the current time, apply the exponential decay function to calculate and generate decay weights;

[0094] In this embodiment, step S33 aims to quantify the timeliness or relevance of earlier interactive behaviors. As time goes by, the reference value of past interactive behaviors for the current behavioral profile usually decreases. By applying an exponential decay function, it can be ensured that the older the interactive behavior, the smaller its weight, thereby reflecting the importance of the immediacy of the behavior to the profile construction. For example, a standard exponential decay function can be used, where the decay coefficient controls the decay rate, and the larger the time interval, the smaller the decay weight. Alternatively, the decay function can also be a piecewise exponential function or a hybrid model combining linear decay to more flexibly adapt to the decay characteristics of different types of interactive behaviors.

[0095] For example, the exponential decay function can be set as follows: In the form of, It is an exponentially decaying weight, and its value range can be set to (0,1]. This is the attenuation coefficient, preset to 0.05, which can be adaptively adjusted according to the task type, such as technical tasks. =0.06, soft skills-related tasks =0.04, It is the time interval between the timestamp of the interaction and the current time, in minutes, and additional constraints can be set: when When ≥60 minutes, The value is fixed at 0.1 to avoid excessive weight decay.

[0096] S34: For the interaction behavior within the preset second time period, calculate and generate enhancement weights based on the density, modification frequency and correlation strength of the interaction behavior within the preset second time period;

[0097] In this embodiment, step S34 aims to identify and highlight high-value interactive behaviors in the recent or critical period. High-density interactions, frequent modifications, and strong correlations with other key behaviors often indicate that the target object has invested more thought, demonstrated a stronger willingness, or exhibited deeper capabilities in this area. By calculating enhancement weights, the impact of these high-value behaviors on the behavioral profile can be amplified. For example, density can be measured by calculating the number of interactive behaviors per unit time; the more behaviors, the higher the density and the greater the weight. Modification frequency can be counted by repeatedly modifying the same information or option within a short period of time; the more modifications, the greater the weight can be adjusted appropriately, specifically increasing or decreasing it based on the information or option. Association strength can be determined by analyzing the semantic or logical connections between interactive behaviors. For example, if a series of interactive behaviors revolve around a specific skill or job requirement, the association strength of these behaviors is high, and the weight can be increased. Furthermore, a multi-factor weighted model can be used, taking density, modification frequency, and association strength as input features, and calculating the comprehensive enhancement weight through linear combination or nonlinear functions.

[0098] S35: Weight the tendency interaction information based on the decay weight and the enhancement weight to generate enhanced interaction information.

[0099] In this embodiment, step S35 aims to apply the various weights calculated previously to the original tendency interaction information to generate enhanced interaction information. By combining the attenuation weights and enhancement weights, a weighted value is assigned to each interaction behavior, thereby generating enhanced interaction information that can more accurately reflect the target object's current behavior pattern and potential intentions. For example, for each interaction behavior, its original information can be multiplied by the calculated corresponding weights or combined in other ways: if the original information is numerical, it can be directly multiplied by the weights; if it is non-numerical, it can be converted into a vector representation and then weighted averaged or weighted summed; or, a weighted feature vector can be constructed, where each dimension represents a feature of an interaction behavior, and corresponding attenuation weights or enhancement weights are applied according to its time period and characteristics.

[0100] Specifically, timestamps of all interactions are extracted, discrete events are organized into a continuous interaction sequence, and this sequence is divided into several manageable time segments based on a preset slice duration. Furthermore, these time segments within the interaction sequence are divided into a preset first time period and a preset second time period. For interactions occurring within the preset first time period, since their occurrence time is relatively early, their reference value for the current behavior profile may decrease over time. Therefore, based on the time interval between the interaction and the current moment, an exponential decay function is applied to calculate and generate decay weights, indicating that the older the interaction, the smaller its impact on the final profile, thus effectively reducing the weight of outdated or less relevant behavior information. Simultaneously, for interactions within the preset second time period, the focus is on identifying their intrinsic value and intensity, by comprehensively considering the interaction's... Dynamic enhancement weights are generated by calculating dimensions such as density, modification frequency, and correlation strength. Specifically, high-density interactions may indicate that the target object is investing more effort in this aspect, frequent modifications may reveal their thoughtful consideration or indecisiveness, while strongly correlated interactions may point to their core interests or abilities. Enhancement weights can amplify the impact of these high-value behaviors on profile construction, making them more prominent in the overall information. Finally, the weights, after attenuation and enhancement processing, are applied to the original tendency interaction information to generate enhanced interaction information. This ensures that the feature extraction stage receives data that has been adjusted for time sensitivity and enhanced in importance, so that the extracted tendency features, ability features, and implicit decision features can more accurately and comprehensively reflect the true situation of the target object, thereby improving the accuracy of personal comprehensive profile construction.

[0101] For example, suppose the target object is completing an interactive task simulating project management. During the interactive task, all its operations are recorded, such as clicking a function button, entering text, dragging elements, and modifying configurations. A precise timestamp is recorded for each operation. The recorded timestamps and corresponding operation content constitute the target object's tendency to interact with the task. First, the timestamps of the interactive behaviors are extracted to form a corresponding interaction sequence. For example, if the task lasts 30 minutes, these 30 minutes may be divided into six consecutive 5-minute time segments. Further, according to a preset time segmentation strategy, the first 20 minutes after the task starts (i.e., the first four time segments) are defined as the preset first time segment, and the last 10 minutes before the task ends (i.e., the last two time segments) are defined as the preset second time segment. For interactive behaviors within the preset first time segment, for example, if the target object performs a preliminary browse of a project module in the second minute after the task starts, an exponential decay is applied based on the interval between the time of this behavior and the current time. The function calculates the decay weight, meaning that if the current moment is the end of the task, the weight of browsing behavior 28 minutes ago will be relatively low. For interactive behaviors within a preset second time period, for example, if the target object modifies the project budget three times within 5 minutes before the task ends and intensively reviews related documents in a short period of time, these behaviors are analyzed. Specifically, the high frequency of budget modification indicates their attention to budget details and decision-making process, while intensive document review reflects their initiative in problem-solving and information acquisition ability. Based on the density, modification frequency, and correlation strength of these behaviors with the core task of budget management, a higher enhancement weight is calculated. Finally, the calculated decay weight and enhancement weight are applied to the original biased interactive information. For example, when generating enhanced interactive information, the description of early browsing behavior will have its importance reduced due to the decay weight, while the description of recent budget modification and document review behavior will have its importance increased due to the enhancement weight, so that the final generated enhanced interactive information can focus more on the target object's recent and high-value interactive behaviors.

[0102] By employing the aforementioned technical solutions, time-series slicing of interactive information and applying differentiated weighting to interactive behaviors at different times, the dynamic behavioral patterns of the target object can be accurately captured. Specifically, exponential decay weighting effectively reduces interference from outdated interactive behaviors, ensuring that the behavioral profile reflects the target object's current tendencies and capabilities, and reducing judgment bias caused by historical information lag. Simultaneously, the dynamic enhancement weighting mechanism highlights recent high-value interactive behaviors, such as frequent modifications, intensive operations, or strongly correlated behaviors. Since these behaviors often better reflect the target object's deeper intentions, decision-making processes, and actual capabilities, the features extracted from the enhanced interactive information are more representative and effective. Through this time-series weighting processing, the accuracy and real-time performance of the constructed comprehensive personal profile can be improved.

[0103] In one embodiment, step S40 includes:

[0104] S41: Input the interactive information and resume text information into the pre-trained feature extraction model, which includes a first extraction layer, a second extraction layer, a third extraction layer and a fusion output layer;

[0105] In this embodiment, the pre-trained feature extraction model is a computational model that has been trained on a large amount of data and is capable of recognizing, parsing, and transforming input data into specific feature representations. Its pre-training process enables it to understand textual information and behavioral data and perform pattern recognition. The feature extraction model can be a model based on a deep learning architecture, such as Transformer, BERT, or their variants. These models are powerful in the field of natural language processing and can capture semantic information and contextual relationships in text. In addition, the feature extraction model can also be a hybrid model that combines traditional machine learning algorithms (such as Support Vector Machine (SVM) and Random Forest) with feature engineering techniques. The feature engineering part is responsible for extracting structured features from the raw data, and then the machine learning algorithm performs pattern learning.

[0106] S42: The first extraction layer extracts skill keywords, project entities, and time series information based on resume text information to form explicit ability features;

[0107] In this embodiment, the first extraction layer is used to identify and extract explicit features directly related to the target object's abilities from the resume text information. The first extraction layer can be implemented using rule-based matching, dictionary matching, and natural language processing techniques to identify skill terms, project names, company names, job titles, and work periods in the resume. Alternatively, it can use a pre-trained word embedding model combined with a convolutional neural network or a recurrent neural network to encode the resume text, thereby automatically learning and extracting vector representations representing skills, projects, and time-series information.

[0108] S43: The second extraction layer extracts preference keywords and value expression patterns based on the feedback content in the preference interaction information;

[0109] In this embodiment, the second extraction layer is used to mine the target object's tendency characteristics, such as its preferences, values, and communication style, from the feedback content in the tendency interaction information. The second extraction layer can use sentiment analysis, topic modeling, or text classification algorithms to process the feedback content and identify positive emotions, negative emotions, focus points, and value orientations implied in the expression methods. Alternatively, it can use a sequence model based on attention mechanisms to analyze the feedback text and capture the target object's deep preferences and potential motivations for specific options, sorting, or text input in different interaction contexts.

[0110] S44: The third extraction layer extracts decision path features and interactive response features based on process data in the tendency interaction information;

[0111] In this embodiment, the third extraction layer is used to extract the implicit decision-making features and interactive response patterns of the target object from the process data in the tendency interaction information. The third extraction layer can use time series analysis and behavior sequence pattern mining algorithms to analyze the timestamps, operation sequences, dwell time and modification history of operation steps to identify feature information such as the degree of hesitation in decision-making, operation proficiency and problem-solving strategies. Alternatively, it can use sequence models such as hidden Markov models or long short-term memory networks to model the process data, thereby capturing the dynamic features of the target object in the interactive task, such as decision-making path, reaction speed and correction behavior.

[0112] Furthermore, the characteristics of the decision-making path include decision hesitation points, modification reversal points, and abandonment points.

[0113] S45: The fusion output layer fuses and vectorizes the outputs of the first extraction layer, the second extraction layer, and the third extraction layer to generate tendency features, ability features, and implicit decision features.

[0114] In this embodiment, the fusion output layer is used to integrate heterogeneous features from different extraction layers and transform them into vectorized representations that can be used for subsequent profile construction. The fusion output layer can employ multimodal fusion techniques, such as attention-based fusion, gating mechanisms, or feature concatenation followed by dimensionality reduction, to ensure the effective combination of features from different sources. Alternatively, a fully connected neural network can be used as the fusion layer, taking the outputs of each extraction layer as input, and learning weights to fuse and output the final vector representations of tendency features, ability features, and implicit decision features.

[0115] Among them, tendency characteristics reflect the preferences, interests, values, and behavioral tendencies exhibited by the target candidate in a specific situation, revealing the target candidate's intrinsic motivation and adaptability in career choices, work environment, teamwork, etc.; ability characteristics reflect the target candidate's knowledge, skills, experience, and professional qualities, which are the basis for measuring whether the target candidate can be competent for a specific position, including hard skills and soft skills demonstrated through project experience; implicit decision-making characteristics refer to the decision-making patterns, thinking habits, problem-solving strategies, and stress resistance exhibited by the target candidate in the interaction process. Usually, implicit decision-making characteristics are difficult to be directly reflected in the resume text, but they are very important for assessing the target candidate's performance and potential in actual work.

[0116] Specifically, the solution in this application achieves multi-dimensional and in-depth feature mining by inputting enhanced interactive information and resume text information into a pre-trained feature extraction model. Specifically, the feature extraction model includes a first extraction layer, a second extraction layer, a third extraction layer, and a fusion output layer. The first extraction layer is used to parse explicit ability features from the resume text information, such as skill keywords, project entities, and time-series information, as a direct reflection of the target candidate's past experience and professional skills. The second extraction layer is used to analyze the feedback content in the interactive information, capturing the target candidate's preference keywords and value expression patterns, thereby revealing the target candidate's inherent career preferences and values. The third extraction layer is used to analyze the process data in the interactive information, extracting decision-making path features and interactive response features. These implicit features can reflect the target candidate's thinking patterns, decision-making styles, and behavioral habits when facing problems. Finally, the fusion output layer integrates and vectorizes the heterogeneous features obtained from these three extraction layers, generating a target feature set containing preference features, ability features, and implicit decision-making features.

[0117] For example, the following process can be adopted when extracting the target feature set: First, the resume text information of the target object and the enhanced interaction information after time-series slicing and weighting are input into a pre-trained feature extraction model. This feature extraction model can be a deep learning model based on the Transformer architecture, with its encoder part used to process text information and its decoder part used to process serialized interaction data. Specifically, the first extraction layer can use the resume text information to identify skill keywords such as "Python programming", "data analysis", and "project management" through named entity recognition technology, identify project entities such as "responsible for a certain project and achieved certain results" through relation extraction, and extract time-series information such as "served as a software engineer from 2018 to 2020" through time-series analysis, which together constitute the explicit ability features of the target object. The second extraction layer targets the feedback content in the tendency interaction information, such as the target object's text answer to a situational question or the ranking result of multiple options in a simulated interview task. It can use a sentiment analysis model to identify the positive or negative tendency in the answer, and use topic modeling to identify the positive or negative tendency in the answer. The first extraction layer identifies the target's focus, such as "teamwork" and "technical challenges," and analyzes their language style, such as "attention to detail" and "results-oriented," thereby extracting preference keywords and value expression patterns. The second extraction layer processes process data in preference interaction information, such as the number of code submissions, the time interval between each submission, the types of errors and modification history during debugging, and the duration of time spent on different modules when the target completes a programming challenge. It can use behavioral sequence analysis algorithms to identify the target's decision path characteristics, such as the number of attempts, error correction efficiency, and decomposition strategies for complex problems, as well as their interactive response characteristics, such as their response speed and modification frequency to task prompts. Finally, the fusion output layer concatenates the capability feature vector output by the first extraction layer, the preference feature vector output by the second extraction layer, and the implicit decision feature vector output by the third extraction layer, and performs dimensionality reduction and nonlinear transformation through a fully connected neural network to generate a unified high-dimensional vector representation. This vector representation contains the target's preference features, capability features, and implicit decision features, thus providing a data foundation for building a comprehensive personal profile.

[0118] Through the above technical solution, this application can deeply mine the key features of the target object from multi-source heterogeneous data. Compared with the existing solution that only relies on resumes or single behavioral data, the hierarchical multimodal feature extraction of this embodiment can not only accurately capture the explicit skills and experience of the target object, but also effectively reveal its inherent career preferences, values, and thinking patterns and decision-making styles shown in actual operation. The resulting comprehensive feature set makes the cognition and understanding of the target object more three-dimensional and accurate, and has the effect of improving the richness and accuracy of the personal comprehensive profile.

[0119] In one embodiment, step S50 includes:

[0120] S51: Calculate the association weights of ability features, tendency features, and implicit decision features through a preset interactive attention mechanism, and adaptively fuse the target feature set based on the calculated association weights to generate an integrated feature vector as a comprehensive personal profile.

[0121] In this embodiment, the preset interactive attention mechanism refers to a computational model used to assign different importance weights among multiple input features. Its function is to simulate the cognitive process of humans giving different levels of attention to different information sources when processing complex information. Through the interactive attention mechanism, the degree of mutual influence between ability features, tendency features, and implicit decision features can be identified and quantified, thereby more accurately understanding the comprehensive characteristics of the target object. Its implementation can include: dynamically generating weights through the calculation of query, key, and value vectors based on the self-attention mechanism in the Transformer architecture; or using a multi-head attention mechanism to capture the correlation between features from different representation subspaces. Calculating the correlation weights of ability features, tendency features, and implicit decision features aims to quantify the degree of interdependence or influence between the various feature dimensions within the target feature set. For example, a person's tendency features may affect the performance of their ability features, while implicit decision features may reflect their behavioral patterns under specific abilities or tendencies. The calculation of correlation weights... This can be achieved by learning the nonlinear relationships between features through neural network models and outputting weights; or by using statistical methods, such as mutual information and Pearson correlation coefficient, to evaluate the linear or nonlinear correlation between features; adaptive fusion of the target feature set based on the calculated association weights refers to dynamically adjusting the contribution ratio of each feature dimension in the comprehensive personal profile according to its association weight, rather than simply linearly superimposing them. This ensures that the characterization of the target object's personal traits in different contexts can highlight the most relevant feature dimensions, improving the accuracy of the profile; its implementation can include: multiplying each feature vector by its corresponding association weight and then performing a weighted sum; or, using a gating mechanism to dynamically control the fusion of information flow according to the weights; generating an integrated feature vector as the comprehensive personal profile. This integrated feature vector is a numerical representation of the comprehensive personal traits of the target object generated after adaptive fusion. It compresses information from multiple dimensions such as ability, tendency, and implicit decision-making into a unified vector space, facilitating analysis and matching.

[0122] S52: Extract hard requirement features, soft requirement features, and implicit cultural features based on job requirement information;

[0123] In this embodiment, hard requirement features, soft requirement features, and cultural implicit features are extracted based on job requirement information. The aim is to identify and quantify different types of job requirements from the original job requirement information. Hard requirement features typically refer to quantifiable and objective skills and experience, such as education, major, years of work experience, and specific technology stacks. Soft requirement features mainly refer to non-technical abilities and qualities, such as communication skills, teamwork, problem-solving skills, and learning abilities. Cultural implicit features typically reflect deep-seated requirements that are difficult to quantify directly, such as corporate culture, team atmosphere, and values, such as innovative spirit, resilience, and adaptability to specific work modes. The extraction methods for these features may include using natural language processing technology to perform entity recognition, keyword extraction, and sentiment analysis on the job description text; or performing semantic matching and feature mapping through a preset rule base and ontology knowledge base.

[0124] S53: Multi-level vector encoding and combination of hard requirements, soft requirements, and cultural implicit characteristics are used to generate a job requirement feature vector with explicit requirements and implicit constraints, which serves as a job requirement profile.

[0125] In this embodiment, multi-layer vector encoding and combination of hard requirement features, soft requirement features, and implicit cultural features refers to converting different types of job requirement features into multi-dimensional numerical vectors, either individually or jointly, and processing them through a multi-layer neural network structure to capture the complex relationships within the features and semantic information at different levels. Combination refers to effectively integrating the encoded feature vectors to form a vector representation that comprehensively represents the job requirements. This can be achieved by: using a multi-layer perceptron to perform nonlinear transformations and combinations on each feature vector; or using a recurrent neural network or Transformer encoder to encode the serialized features and ultimately generate a fixed-length vector. This generates a job requirement feature vector with explicit requirements and implicit constraints, serving as a job requirement profile. This feature vector, generated after multi-layer encoding and combination, accurately represents the numerical representation of the target company's job requirements. It includes not only explicit requirements such as skills and experience but also implicit constraints such as candidates' soft qualities and cultural fit.

[0126] Specifically, when constructing a comprehensive profile of a target individual, the first step is to analyze the ability features, tendency features, and implicit decision-making features extracted from enhanced interactive information and resume text information using a pre-defined interactive attention mechanism. This interactive attention mechanism helps identify and quantify the intrinsic relationships between these different dimensional features. For example, a person's specific ability may be closely related to a certain tendency in their interactive tasks, while their implicit decision-making features may further reveal the deeper driving factors of their abilities and tendencies. By calculating the correlation weights of these features, their relative importance and mutual influence in characterizing the comprehensive traits of the target individual can be understood. Furthermore, the target feature set is adaptively adjusted based on the calculated correlation weights. This process involves fusion to generate a more accurate integrated feature vector, serving as a comprehensive profile of the target individual. Simultaneously, when constructing a profile of the target company's job requirements, it first extracts traditional hard requirement features based on job requirement information, while further identifying and quantifying soft requirement features and implicit cultural features. The extracted features are then subjected to multi-layer vector encoding and combination. Multi-layer encoding captures the complex semantics and information at different levels of abstraction within each feature, while combination effectively integrates the encoded features to generate a job requirement feature vector with explicit requirements and implicit constraints. This job requirement feature vector, as a job requirement profile, can more comprehensively reflect the company's expectations for candidates, including cultural and soft requirements that are difficult to articulate directly.

[0127] For example, suppose the target candidate is looking for a software development engineer position, and the target company is recruiting a senior Java development engineer. When constructing the target candidate's comprehensive profile, the first step is to extract ability characteristics (such as Java programming skills and data structure knowledge), tendency characteristics (such as a preference for independent work and enthusiasm for new technologies), and implicit decision-making characteristics (such as a tendency to conduct comprehensive analysis before coding when solving complex problems) from their resume text information and interactive tasks. To more accurately characterize the target candidate, a pre-defined interactive attention mechanism is activated. For example, a Transformer-based attention network can be used to analyze the target candidate's response patterns to different technical questions in interactive tasks. If the target candidate demonstrates high focus and in-depth thinking in interactive tasks involving Java concurrent programming, and their resume also reflects rich relevant project experience, then the interactive attention mechanism may calculate a high correlation weight between "Java programming skills" and "enthusiasm for new technologies." Simultaneously, the implicit decision-making characteristic of "a tendency to conduct comprehensive analysis before solving complex problems" is also closely related to these abilities and tendencies. Based on the calculated correlation weights, the target candidate's ability characteristics, tendency characteristics, and implicit decision-making characteristics are adaptively fused. For example, if "Java programming skills..." The highest weight is given to "passionate about new technologies" and "analytical decision-making," which will be given greater weight during the integration process. At the same time, related characteristics such as "passionate about new technologies" and "analytical decision-making" will be incorporated with appropriate weights to generate an integrated feature vector. This integrated feature vector comprehensively represents the overall characteristics of the target candidate as a senior Java development engineer. Simultaneously, when constructing the job requirement profile of the target company, hard requirement characteristics (such as 5+ years of Java development experience, familiarity with Spring Cloud microservice architecture, bachelor's degree or above) and soft requirement characteristics (such as excellent teamwork skills, good communication skills) are extracted from the "Senior Java Development Engineer" job requirements information published by the company. The system identifies key characteristics such as the ability to solve complex problems and implicit cultural traits (e.g., adaptability to a fast-paced work environment, innovative spirit, and alignment with the company's "customer-first" value). These characteristics are then encoded and combined using multi-layered vectors. For example, hard requirements can be encoded using word embeddings and convolutional neural networks, soft requirements can be semantically encoded using pre-trained language models, and implicit cultural traits can be encoded through analysis of corporate culture descriptions and sentiment analysis. The encoded feature vectors are then combined using a multi-layered perceptron to generate a job requirement feature vector with explicit requirements and implicit constraints, serving as a profile of the target company's requirements for that position.

[0128] Through the above technical solutions, this application can achieve a deep understanding of the personal characteristics of the target object and the job requirements of the target enterprise. Specifically, in terms of personal profile construction, by introducing an interactive attention mechanism and adaptive fusion, the correlation between abilities, tendencies, and implicit decision-making characteristics is captured, so that the comprehensive personal profile not only includes explicit information, but also combines behavioral patterns and potential preferences, thereby achieving a comprehensive and accurate portrayal of the target object. In terms of job profile construction, by extracting hard, soft, and cultural implicit characteristics and performing multi-layer vector encoding and combination, a job requirement feature vector that includes both explicit skill requirements and implicit cultural fit can be generated, making the job profile more three-dimensional and realistic, and improving the accuracy of matching between individuals and jobs.

[0129] In one embodiment, the analysis model includes a mapping comparison layer, a parallel computing layer, and a score generation layer. Step S60 includes:

[0130] S61: The mapping and comparison layer performs bidirectional mapping and consistency comparison analysis on the comprehensive personal profile and the job requirement profile, generating several feature pairs. The feature pairs include ability features and hard requirement features, tendency features and soft requirement features, and tendency features and cultural implicit features.

[0131] In this embodiment, the analysis model refers to a structured processing framework used to match and evaluate the input comprehensive personal profile and job requirement profile to generate an adaptability score. The analysis model includes a mapping and comparison layer, a parallel computing layer, and a score generation layer. The mapping and comparison layer is a module layer in the analysis model, used to identify and establish the correspondence between different dimensions of features between the personal profile and the job profile. Specifically, the mapping and comparison layer ensures that each feature in the personal profile can be effectively compared with the corresponding requirements in the job profile by performing bidirectional mapping and consistency comparison analysis. For example, the mapping and comparison layer can use semantic matching algorithms to identify similarities in text descriptions, or use predefined ontology to associate similar features under different expressions. Furthermore, the degree of association between features can be quantified by calculating the similarity of feature vectors. Further, several feature pairs output by the mapping and comparison layer, such as ability features and hard requirement features, tendency features and soft requirement features, and tendency features and cultural implicit features, can provide explicit evaluation units for the parallel computing layer.

[0132] S62: The parallel computing layer calculates the initial fit of each feature pair in parallel, and corrects the calculated initial fit by using a prior matrix based on historical collaborative data to generate a corrected fit.

[0133] In this embodiment, the parallel computing layer is a module layer in the analysis model. After receiving the feature pairs generated by the mapping and comparison layer, it independently calculates the initial fit for each feature pair. Its parallel processing improves computational efficiency and allows for the use of the most suitable fit calculation method for different types of feature pairs. For example, for matching ability features with hard requirement features, exact matching or fuzzy matching algorithms can be used; for matching preference features with soft requirement features, prediction based on machine learning models or similarity measurement methods can be used. Furthermore, the parallel computing layer also corrects the initial fit by introducing a prior matrix based on historical collaborative data. This prior matrix can store the weights or correction factors of different feature pairs in past successful matching cases, making the fit calculation closer to the actual recruitment effect. For example, if historical data shows that a certain ability is particularly important for job success, even if the initial fit is average, it can be appropriately improved through the prior matrix.

[0134] S63: The scoring generation layer calculates and generates an adaptive score based on the modified fit.

[0135] In this embodiment, the score generation layer is a module layer in the analysis model, used to synthesize all the corrected fits output by the parallel computing layer and calculate the final fitness score accordingly. The score generation layer can adopt various aggregation strategies, such as linearly combining all corrected fits by weighted summation, where the weights can be preset according to the importance of the feature pairs or learned through a machine learning model. Alternatively, a complex nonlinear model, such as a neural network model, can be used to take the corrected fits as input and output the final fitness score.

[0136] Specifically, the proposed solution refines the analysis model into a multi-layered architecture comprising a mapping and comparison layer, a parallel computing layer, and a score generation layer. This enables a structured evaluation of the matching process between an individual's comprehensive profile and the job's requirement profile. Specifically, the mapping and comparison layer first ensures the correspondence of multi-dimensional features between the individual and the job, avoiding information omissions or mismatches. The parallel computing layer performs parallel and independent fit assessments on the feature pairs output by the mapping and comparison layer. By introducing a prior matrix based on historical collaborative data, past successful experiences can be incorporated into the current matching judgment, making the fit calculation more practically instructive and predictive. Finally, the score generation layer aggregates the corrected fit scores to generate accurate and interpretable adaptive scores.

[0137] For example, as a specific implementation method, when the comprehensive personal profile and the job requirement profile are input into the analysis model, the mapping and comparison layer performs comparative analysis on them. For instance, for the ability characteristics in the comprehensive personal profile (such as "proficient in Python programming" and "possessing project management experience") and the hard requirement characteristics in the job requirement profile (such as "requiring more than 3 years of Python development experience" and "PMP certificate preferred"), the mapping and comparison layer will use natural language processing technology and knowledge graph matching to identify the correspondence between "Python programming" and "Python development experience," as well as the correlation between "project management experience" and "PMP certificate," and generate corresponding feature pairs. At the same time, for preference characteristics (such as "preferring teamwork" and "emphasizing innovation") and soft requirement characteristics (such as "possessing teamwork spirit" and "having innovative awareness"), as well as preference characteristics and implicit cultural characteristics (such as "adapting to a fast-paced work rhythm" and "identifying with a flat management culture"), the mapping and comparison layer can utilize semantic similarity calculation or pre-training. The system trains a classifier to establish correspondences between these features and forms their respective feature pairs. Subsequently, the parallel computing layer calculates the initial fit of the feature pairs output by the mapping and comparison layer. In the parallel computing layer, the initial fit of each feature pair is calculated independently. For example, the fit between "Python programming ability" and "Python development experience" may be obtained by comparing skill level and years of experience to arrive at an initial score. At the same time, the initial fit is corrected by a prior matrix constructed based on historical collaborative data. If historical data shows that the fit of "Python programming ability" has a particularly large impact on the final success rate for this position, then even if the initial fit is slightly low, the prior matrix may be adjusted upwards to reflect its importance. After the corrected fit of all feature pairs is calculated, the scoring generation layer can use a multilayer perceptron network, taking all corrected fits as input, and using the learned weights and activation functions, finally outputting an adaptive score. This adaptive score can intuitively reflect the overall fit between the target object and the target position.

[0138] Through the above technical solutions, the multi-layered analysis model framework provided in this application can not only comprehensively consider the multi-dimensional fit between individuals and positions, but also make the scoring results more practically instructive by introducing historical collaborative data for correction, thereby improving the accuracy and interpretability of the adaptability score.

[0139] Furthermore, all preset parameters in this application (such as attenuation coefficient, matching threshold, weight coefficient, model dimension, etc.) are basic preset values, which can be adaptively adjusted according to the recruitment field (such as Internet, manufacturing, finance), job type (technical position, management position, operation position), enterprise size (large enterprise, small and medium-sized enterprise) and actual needs. The adjusted parameters are still within the protection scope of this application, and the adjustment process does not require changing the core technical solution of this application.

[0140] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0141] In one embodiment, a behavioral profiling-based AI-powered intelligent recommendation system for enterprise recruitment is provided. This system corresponds one-to-one with the behavioral profiling-based AI-powered intelligent recommendation method for enterprise recruitment described in the previous embodiment. The behavioral profiling-based AI-powered intelligent recommendation system for enterprise recruitment includes:

[0142] The information acquisition module is used to acquire the resume text information of the target object, and based on the resume text information, to acquire the job requirements information of the target company.

[0143] The task generation module is used to generate interactive tasks based on resume text information and job requirement information and provide them to the target object, and to collect the target object's tendency interaction information based on the interactive tasks. The tendency interaction information includes feedback content and process data collected through the interactive tasks.

[0144] The information enhancement module is used to perform time-series slicing of the interactive information according to timestamps, set exponential decay weights for interactive behaviors in a preset first time period, set dynamic enhancement weights for interactive behaviors in a preset second time period, and generate enhanced interactive information.

[0145] The feature extraction module is used to extract a target feature set based on enhanced interaction information and resume text information. The target feature set includes tendency features, ability features, and implicit decision features.

[0146] The profile building module is used to build a comprehensive personal profile of the target object based on the target feature set, and to build a job requirement profile of the target company based on job requirement information;

[0147] The scoring generation module is used to input the individual's comprehensive profile and the job requirement profile into the pre-trained analysis model to generate an adaptive score;

[0148] The information generation module is used to generate recommendation information including adaptation points based on the adaptability score, and send the recommendation information as matching result information to the target enterprise and the target object respectively.

[0149] For specific limitations regarding the AI-powered intelligent recommendation system for enterprise recruitment based on behavioral profiling, please refer to the limitations of the AI-powered intelligent recommendation method for enterprise recruitment based on behavioral profiling mentioned above, which will not be repeated here. Each module in the aforementioned AI-powered intelligent recommendation system for enterprise recruitment based on behavioral profiling can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0150] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements an AI-powered intelligent recommendation method for enterprise recruitment based on behavioral profiles.

[0151] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements an AI-powered intelligent recommendation method for enterprise recruitment based on behavioral profiles.

[0152] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for AI-powered intelligent recommendation in enterprise recruitment based on behavioral profiles, characterized in that: Including the following steps: Obtain the resume text information of the target object, and obtain the job requirements information of the target company based on the resume text information; An interactive task is generated based on the resume text information and job requirements information and provided to the target object. The target object's tendency interaction information based on the interactive task is collected. The tendency interaction information includes feedback content and process data collected through the interactive task. The interactive information is sliced ​​according to the timestamp, and the interactive behavior in the first preset time period is given an exponential decay weight, and the interactive behavior in the second preset time period is given a dynamic enhancement weight to generate enhanced interactive information. A target feature set is extracted based on enhanced interactive information and resume text information. The target feature set includes tendency features, ability features, and implicit decision features. A comprehensive personal profile of the target object is constructed based on the target feature set, and a job requirement profile of the target company is constructed based on the job requirement information. Input the individual's comprehensive profile and the job requirement profile into the pre-trained analysis model to generate an adaptive score; Recommendations are generated based on the adaptability score, including key adaptability points. These recommendations are then sent as matching results to the target company and the target individual.

2. The AI-powered intelligent recommendation method for enterprise recruitment based on behavioral profiling as described in claim 1, characterized in that: The steps of obtaining the resume text information of the target object and obtaining the corresponding job requirement information of the target company based on the resume text information include: Obtain resume text information, which includes skills text information, experience text information, and intended job text information; Keyword extraction and intent parsing are performed on the text information of intended job positions to generate a set of job intentions; Based on the set of job preferences, a matching search is performed in the pre-set job information database to filter out the target company set and the corresponding job requirements.

3. The AI-powered intelligent recommendation method for enterprise recruitment based on behavioral profiling as described in claim 1, characterized in that: The step of generating an interactive task based on resume text information and job requirement information and providing it to the target audience, and collecting the target audience's preferred interaction information based on the interactive task, including feedback content and process data collected through the interactive task, includes: Analyze job requirements information to extract core competency requirements and work context elements; Based on the resume text information, core competency requirements, and work context elements, a multi-dimensional match is performed in the pre-set task template library to select several related basic task templates. According to the preset evaluation logic rules, the selected basic task templates are sorted and logically linked to generate interactive tasks. Interactive tasks are provided to the target object, and the target object's interaction log is recorded to form a biased interaction information.

4. The AI-powered intelligent recommendation method for enterprise recruitment based on behavioral profiling as described in claim 1, characterized in that: The steps of slicing the tendency interaction information according to timestamps, setting exponential decay weights for interaction behaviors within a preset first time period, setting dynamic enhancement weights for interaction behaviors within a preset second time period, and generating enhanced interaction information include: Extract the timestamps of all interactive behaviors in the tendency interaction information to form an interaction sequence, and divide the interaction sequence into several consecutive time segments based on the preset slice duration. Based on a preset time segmentation strategy, a preset first time segment and a preset second time segment are determined within the time segment; For interactive behaviors within the preset first time period, based on the interval between their timestamps and the current time, an exponential decay function is applied to calculate and generate decay weights. For interactive behaviors within the preset second time period, enhancement weights are calculated and generated based on the density, modification frequency, and correlation strength of the interactive behaviors within the preset second time period. Enhanced interaction information is generated by weighting the tendency interaction information based on decay weights and enhancement weights.

5. The AI-powered intelligent recommendation method for enterprise recruitment based on behavioral profiling as described in claim 1, characterized in that: The step of extracting a target feature set based on enhanced interactive information and resume text information, wherein the target feature set includes tendency features, ability features, and implicit decision features, includes: The interactive information and resume text information are input into a pre-trained feature extraction model, which includes a first extraction layer, a second extraction layer, a third extraction layer, and a fusion output layer. The first extraction layer extracts skill keywords, project entities, and time series information based on resume text information to form explicit ability features; The second extraction layer extracts preference keywords and value expression patterns based on the feedback content in the preference interaction information. The third extraction layer extracts decision path features and interactive response features based on process data in the tendency interaction information; The fusion output layer fuses and vectorizes the outputs of the first extraction layer, the second extraction layer, and the third extraction layer to generate tendency features, ability features, and implicit decision features.

6. The AI-powered intelligent recommendation method for enterprise recruitment based on behavioral profiling as described in claim 1, characterized in that: The steps of constructing a comprehensive personal profile of the target object based on the target feature set and constructing a job requirement profile of the target enterprise based on job requirement information include: The association weights of ability features, tendency features, and implicit decision features are calculated through a preset interactive attention mechanism. Based on the calculated association weights, the target feature set is adaptively fused to generate an integrated feature vector as a comprehensive personal profile. Based on job requirements information, extract hard requirements features, soft requirements features, and implicit cultural features; By using multi-layered vector encoding and combination of hard requirements, soft requirements, and implicit cultural characteristics, a job requirement feature vector with explicit requirements and implicit constraints is generated, serving as a job requirement profile.

7. The AI-powered intelligent recommendation method for enterprise recruitment based on behavioral profiling as described in claim 6, characterized in that: The analysis model includes a mapping and comparison layer, a parallel computing layer, and a score generation layer. The step of inputting the individual comprehensive profile and the job requirement profile into the pre-trained analysis model to generate an adaptive score includes: The mapping and comparison layer performs bidirectional mapping and consistency comparison analysis between the individual comprehensive profile and the job requirement profile, generating several feature pairs. These feature pairs include ability features and hard requirement features, tendency features and soft requirement features, and tendency features and cultural implicit features. The parallel computing layer calculates the initial fit of each feature pair in parallel, and corrects the calculated initial fit by using a prior matrix based on historical collaborative data to generate a corrected fit. The scoring generation layer calculates and generates an adaptive score based on the modified fit.

8. A corporate recruitment AI intelligent recommendation system based on behavioral profiles, characterized in that, include: The information acquisition module is used to acquire the resume text information of the target object, and based on the resume text information, to acquire the job requirements information of the target company. The task generation module is used to generate interactive tasks based on resume text information and job requirement information and provide them to the target object, and to collect the target object's tendency interaction information based on the interactive tasks. The tendency interaction information includes feedback content and process data collected through the interactive tasks. The information enhancement module is used to perform time-series slicing of the interactive information according to timestamps, set exponential decay weights for interactive behaviors in a preset first time period, set dynamic enhancement weights for interactive behaviors in a preset second time period, and generate enhanced interactive information. The feature extraction module is used to extract a target feature set based on enhanced interaction information and resume text information. The target feature set includes tendency features, ability features, and implicit decision features. The profile building module is used to build a comprehensive personal profile of the target object based on the target feature set, and to build a job requirement profile of the target company based on job requirement information; The scoring generation module is used to input the individual's comprehensive profile and the job requirement profile into the pre-trained analysis model to generate an adaptive score; The information generation module is used to generate recommendation information including adaptation points based on the adaptability score, and send the recommendation information as matching result information to the target enterprise and the target object respectively.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the AI-powered intelligent recommendation method for enterprise recruitment based on behavioral profiles as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the AI-powered intelligent recommendation method for enterprise recruitment based on behavioral profiling as described in any one of claims 1-7.

Images