An AI-driven precise matching system for employment supply and demand targeting key groups

Abstract

This invention discloses an AI-driven precise matching system for employment supply and demand targeting key groups, specifically in the field of employment service data processing. It addresses the problem of conditional job-seeking expressions failing to accurately match job positions. The system acquires job seekers' interactive text expressing their employment intentions, identifies the expressing subject, segments semantic fragments, and identifies conditional expressions, generating a candidate set of job-seeking constraints that includes the constraint object, intention orientation, conditions for validity, and source location. The system then groups these constraints by object and performs constraint synthesis and discrimination to generate a job-seeking condition constraint graph. Subsequently, it performs condition normalization processing and extracts performance features from job information to generate a job constraint judgment table. Finally, based on this, it filters, displays, and sorts the job opportunities to be matched, outputting job recommendation results that include recommendation order, constraint basis, and condition prompts.

Description

Technical Field

[0001] This invention relates to the field of employment service data processing, and more specifically, to an AI-driven precise matching system for employment supply and demand targeting key groups. Background Technology

[0002] In the digitalization of public employment services, matching employment supply and demand for key groups typically requires obtaining job seekers' employment intentions through online forms, in-person Q&A sessions, telephone follow-ups, or voice transcription, and then converting these intentions into data suitable for job matching. The employment choices of key group job seekers are often influenced by various factors such as working hours, commuting distance, accommodation, salary arrangements, physical endurance, and family care arrangements. Their employment intentions are not always simply expressed as acceptance or rejection, but often appear in the form of conditions, concessions, exceptions, supplementary explanations, or subsequent modifications. For example, job seekers might express that they are willing to work day shifts but not night shifts, are willing to accept long commutes if accommodation is provided, are willing to accept lower wages but cannot accept the lack of social security, and are willing to temporarily fill in for others but not long-term night shifts. Such expressions simultaneously contain the subject of the constraint, the orientation of the intention, the prerequisites for effectiveness, and the order of appearance. If these cannot be fully identified, it will be difficult to accurately reflect the true employment boundaries of job seekers.

[0003] Current employment service data processing typically focuses on job seeker profiling, job profiling, tag extraction, information retrieval, and recommendation ranking. A common practice is to break down interactive text into static tags and then match these tags with job field information. While this approach meets general job recommendation needs, it often lacks a fine-grained relationship reconstruction mechanism for situations where there are discrepancies in expression, changes in preconditions, or exceptions within the same constraint. Because the conditions, intentions, and sources within the interactive text are not uniformly organized, the system may misinterpret expressions acceptable under specific conditions as unconditionally acceptable, directly identify subsequent supplementary explanations as contradictory, or incorrectly merge opposing intentions within the same context. These processing biases, when applied to job matching, allow jobs that do not meet actual employment constraints to continue to be included in the recommendation pool, increasing redundant communication between job seekers and employment service personnel, and making it difficult for job recommendations to reflect the actual achievable employment conditions for key groups. Therefore, there is an urgent need for a precise employment supply and demand matching solution that can extract conditional job-seeking constraints from interactive text, reconstruct conflicting and corrective relationships within the same constraint, and consistently match the resulting constraint relationships with job information.

[0004] To address the aforementioned problems, a technical solution is provided. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an AI-driven precise matching system for employment supply and demand targeting key groups. This system acquires job seekers' interactive text expressing their employment intentions, performs subject identification, semantic segmentation, and conditional expression recognition, generating a candidate set of job constraint criteria that includes the constraint object, intention orientation, conditions for validity, and source location. The system then groups these criteria by constraint object and performs constraint synthesis and discrimination to generate a job constraint graph. Subsequently, it performs condition normalization processing and performance feature extraction on job information to generate a job constraint judgment table. Finally, based on this, it filters, displays, and sorts the job opportunities to be matched, outputting job recommendation results that include recommendation order, constraint basis, and condition prompts, thereby solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: An AI-driven precise matching system for employment supply and demand targeting key groups includes a willingness extraction module, a constraint graph construction module, a job determination module, and a recommendation control module. The intention extraction module obtains interactive text that represents job seekers' employment intentions, performs subject identification, semantic segmentation, and conditional expression identification on the interactive text, and generates a candidate set of job application constraints that includes the constraint object, intention orientation, conditions for establishment, and source location. The constraint graphing module receives a set of job application constraint candidates, groups the set of job application constraint candidates according to the constraint objects, and performs constraint synthesis and discrimination based on the compatibility of the conditions, the opposite relationship of the intention orientation, and the order of the source location to generate a job application constraint graph. The job determination module obtains job information for potential employment opportunities and matches them with job seeker requirements. Figure 1 The constraint expression system is subjected to condition normalization and performance feature extraction to generate a job constraint judgment table. The recommendation control module filters, displays, and sorts the job opportunities to be matched based on the job constraint judgment table, and outputs job recommendation results that include the recommendation order, constraint basis, and condition prompts.

[0007] Furthermore, the data processing server performs basic text normalization on the interactive text, processing sentence segmentation markers, pause markers, and transcription noise markers into identifiable sentence boundaries, and forming an interactive text sequence according to the time of interaction. The subject of expression identification determines the job seeker's expression of employment intention based on speaker identifiers, title references, employment intention predicates, and job description predicates, retains the rounds and character ranges of non-job seeker employment intention expressions as source location references, and sends the text belonging to the job seeker's employment intention expression into semantic segmentation.

[0008] Furthermore, semantic fragment segmentation is performed by forming initial fragments based on sentence boundaries, and then further segmented according to the dominance relationships of conditional conjunctions, concession conjunctions, and exception prompts. The data processing server forms a dependency relationship sequence based on conditional triggers, concession triggers, exception triggers, negation triggers, and intention triggers, and determines the constraint object, intention orientation, conditions for establishment, and source location from the dependency relationship sequence. The constraint content of each semantic fragment is then written into the job application constraint candidate set in a unified writing order.

[0009] Furthermore, the job constraint candidate set is grouped according to the constraint object naming rules in the constraint expression system. Job constraint candidates with the same constraint object normalization result are grouped into the same candidate group. The data processing server sorts the job constraint candidates in the candidate group according to the source location to form a constraint expression sequence. The item words that do not match the constraint object naming rules are written into the constraint objects to be confirmed, so that the confirmed constraint objects can continue to participate in the grouping and constraint relationship reconstruction process.

[0010] Furthermore, the constraint synthesis discrimination includes the calculation of contextual conflict. The data processing server forms a candidate pairing sequence within the same candidate group and compares the establishment condition relationship and intention orientation relationship of each candidate pairing. When the establishment condition relationship is the same, covering, or compatible, and the intention orientation is acceptance and exclusion, conditional acceptance and exclusion, or acceptance and exclusion under the same establishment condition, it is included in the contextual conflict record, and a contextual conflict is formed based on the contextual conflict record and the number of comparable candidate pairings.

[0011] Furthermore, the constraint synthesis discrimination also includes ordinal correction calculation and hierarchical evidence reasoning. The data processing server forms a sequential pairing sequence according to the source position, generates a corrected establishment record based on the subsequent source position, changes in intention orientation and narrowing relationship of establishment conditions, and outputs conventional merging, overlay resolution or parallel retention according to the same context conflict and ordinal correction, and controls the generation of main nodes, condition nodes and edge attributes of job application condition constraint graph, and retains the corresponding source position as constraint basis data.

[0012] Furthermore, the job information is aligned with constraint objects through a constraint expression system, which includes constraint object naming rules, establishment condition normalization rules, and performance attribute expression rules. The data processing server first divides the job information into job description fragments, then assigns the job items in the job description fragments to constraint objects consistent with the main nodes of the job application condition constraint diagram, and forms a sequence of job description fragments organized by constraint objects, which is used as input for condition normalization processing and performance feature extraction.

[0013] Furthermore, performance feature extraction transforms job performance content into constraints related to job application requirements. Figure 1The conditions are expressed and written into the job performance feature set; the data processing server writes the performance conditions that the job has already met, the performance conditions that the job has not met, and the performance conditions that the job has not specified, according to different situations: the job information clearly states the conditions, the opposite facts are clearly recorded, and only the facts that need further confirmation are recorded. The corresponding constraint objects, job performance content, and condition expression results are retained.

[0014] Furthermore, the job constraint judgment table is generated based on the matching results between the job performance feature set and the job application condition constraint graph. The data processing server matches the job's existing performance conditions, unmet performance conditions, and undefined performance conditions with the valid constraint edges in the job application condition constraint graph, and writes them into the satisfied state, condition satisfied state, and unsatisfied state according to the different intention orientations of acceptance, exclusion, and conditional acceptance. At the same time, the corresponding conditions for establishment are retained. The valid constraint edges are the constraint edges that are retained and participate in the matching after constraint synthesis and discrimination.

[0015] Furthermore, the data processing server performs recommendation control based on the job constraint judgment table, writes unmet job opportunities into the filtering candidate queue, writes unmet job opportunities into the condition display candidate queue, writes job opportunities into the condition display candidate queue where all valid constraint edges are met, and writes unmet job opportunities into the sorting candidate queue. The server then generates a recommendation order based on the number of met states, the number of condition met states, the number of fulfillment conditions already met by the job, and the job information set entry order.

[0016] The technical effects and advantages of this invention, which describes an AI-driven precise matching system for employment supply and demand targeting key groups: This invention transforms job seekers' employment intentions in interactive text into a candidate set of job constraint elements, including the constraint object, intention orientation, conditions for validity, and source location. This allows conditional expressions, concession expressions, exception expressions, and modification expressions to participate in data processing with a unified structure, reducing the loss of intention boundaries caused by relying solely on keyword tags. By preserving the source location, it is possible to distinguish the sequential expression relationships under the same constraint object, providing a calculable processing basis for supplementary explanations, exceptions, and conditional premises in the text.

[0017] This invention uses constraint synthesis and discrimination based on conditional compatibility, opposite intention orientation, and source position order to reconstruct job application constraint candidates under the same constraint object into a job application condition constraint graph. This process ensures that acceptance, exclusion, and conditional acceptance no longer exist as separate labels, but rather express their logical relationships in the form of master nodes, condition nodes, and edge attributes. This reduces the risk of erroneously merging opposite intentions under the same conditional context, and also reduces the risk of incorrectly retaining the expression of subsequent source position corrections as parallel contradictions.

[0018] This invention is based on job application requirements. Figure 1 The system's constraint expression mechanism processes job information, generates a job constraint judgment table, and categorizes job opportunities to be matched into different recommendation control states, such as filtering, condition display, and sorting. The output includes the recommendation order, constraint basis, and condition hints, ensuring that job recommendations not only present a job ranking but also retain the source of constraints and condition states that formed the recommendation control. This helps reduce the number of jobs that do not meet the job search constraints from entering the recommendation display range. Attached Figure Description

[0019] Figure 1 This is a diagram illustrating the overall architecture of an AI-driven precise matching system for employment supply and demand targeting key groups, as described in this invention. Figure 2 This is a flowchart of the job seeker employment intention extraction process in this invention; Figure 3 This is a schematic diagram illustrating the generation of job application condition constraint diagrams in this invention; Figure 4 This is a flowchart of the job constraint determination and recommendation control process in this invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figure 1 - Figure 4 This invention provides an AI-driven precise matching system for employment supply and demand targeting key groups, comprising: This invention processes interactive text for employment services and job information for matching job opportunities. First, a willingness extraction module identifies job seekers' expressed employment intentions from the interactive text, segments them into semantic fragments, and extracts the constraint objects, intention orientations, conditions for fulfillment, and source locations related to job matching, forming a candidate set of job constraint elements. Next, a constraint graphing module groups the candidate set of job constraint elements according to the constraint objects, and combines the compatibility of conditions for fulfillment, the inverse relationship of intention orientations, and the order of source locations to perform constraint synthesis and discrimination, generating a job constraint diagram expressing the logical relationship between constraint objects, conditions for fulfillment, and intention orientations. Subsequently, a job determination module performs condition normalization processing and extracts performance features from the job information according to the constraint expression system, forming a job performance feature set and generating a job constraint determination table. Finally, a recommendation control module filters, displays, and sorts the job opportunities to be matched based on the job constraint determination table, generating a job recommendation result that includes recommendation order, constraint basis, and condition prompts.

[0022] All of the following modules are executed by the data processing server. In this embodiment, AI-driven means that the data processing server uses natural language processing rules, constraint expression system and hierarchical evidence reasoning model to perform structured recognition, constraint relationship reconstruction and job performance matching of conditional job-seeking expressions in employment service interaction text; this process does not require training samples as a necessary condition and can complete the employment supply and demand matching control through preset rule set, source position order and evidence reasoning judgment conditions.

[0023] The constraint expression system is pre-configured in the data processing server as a set of rules, and the same set of rules is used when generating the job application constraint candidate set, the job application condition constraint graph, and the job performance feature set. This set of rules includes constraint object naming rules, condition unification rules, and performance attribute expression rules. The constraint object naming rules use the employment service field standard as the primary key and are extended by synonym expression mapping rules to classify different natural language expressions into a unified constraint object. The condition unification rules are used to break down condition content into condition objects, condition attributes, and condition values. The performance attribute expression rules are used to write job facts as fulfilled, unmet, or unclear. Condition objects specify the employment matter being judged, condition attributes specify the judgment dimension, and condition values ​​specify specific requirements or factual content. For example, night shift, day shift, rotating shifts, and temporary shift work are classified as working hours; proximity to home, long commute, 1-hour commute, and 12-kilometer distance are classified as commuting distance; accommodation provided, dormitory provided, employee dormitory, and accommodation application are classified as accommodation conditions; wages, salaries, base salary, social security, and subsidies are classified as compensation arrangements. If a term in a semantic fragment or job description fragment does not match the naming rules for constraint objects, the data processing server retains the original expression and source location of the term and writes it into the constraint object to be confirmed. After the employment service personnel confirm the corresponding constraint object, the data processing server writes the confirmation result into the constraint expression system and continues to perform grouping, normalization, and matching judgments according to the confirmed constraint object. Providing employee dormitories is normalized to accommodation conditions, having, or providing; temporary shift work is normalized to working hours, temporary, or shift work; commuting within 1 hour is normalized to commuting distance, duration, or not exceeding 1 hour. When the job information clearly specifies the condition value, it is written as "already met"; when the job information clearly specifies the opposite of the condition value, it is written as "not met"; when the job information only indicates that it can be applied for, is negotiable, is negotiable, or requires further confirmation, it is written as "not specified".

[0024] The intention extraction module obtains interactive text that represents job seekers' employment intentions, identifies the subject of expression, segments semantic fragments, and identifies conditional expressions in the interactive text, and generates a candidate set of job application constraints that includes the constraint object, intention orientation, conditions for establishment, and source location.

[0025] The intention extraction module takes the interactive text generated during the employment service interaction as the starting point for processing. The interactive text comes from the question-and-answer records between job seekers and employment service personnel, speech-to-text transcripts, or supplementary instructions filled in online. The data processing server forms an interactive text sequence according to the time of the interaction and retains the sequential position of each round of text in the sequence, so that the subsequent source position can be traced back to the specific semantic fragment.

[0026] The data processing server first performs basic text reorganization on the interactive text, processing sentence breaks, pauses, and transcription noise markers from the same round of interaction into recognizable sentence boundaries without changing the original word order; then, it performs subject identification. Subject identification combines speaker identifier matching with semantic role determination. The data processing server first generates a subject determination record for each text segment in the interactive text. The subject determination record is jointly determined by the speaker identifier, title, employment intention predicate, and job description predicate. When the speaker identifier is a job seeker, the text segment is directly written as the job seeker's employment intention expression; when the speaker identifier is missing, the data processing server searches for titles related to the job seeker's own state, such as "I," "myself," "family," "body," "children," and "elderly," and searches for employment intention predicates such as "can," "cannot," "inconvenient," "willing," "accept," "consider," and "not consider" in the text segment; when the title and employment intention predicate are in the same semantic segment, and the semantic segment does not use job description predicates such as "job offering," "company arrangement," or "recruitment requirements" as the subject-predicate structure, the text segment is determined to be the job seeker's employment intention expression. When a text fragment only contains job responsibilities, recruitment conditions, or questions from service personnel, the job seeker's expression of employment intention is not included, but the round number and character range are retained as a reference for the source location.

[0027] In practice, if an employment service worker asks if a job seeker can accept night shifts, and the job seeker replies that they would consider it if it weren't consecutive night shifts, as they need to take care of elderly relatives at night, the data processing server will identify the inquiry as a guiding expression and the answer as an expression of the job seeker's employment intention. The server will also retain the order of the responses, the character ranges within the sentences, and the order of the interaction times. After the subject of expression is identified, text that represents the job seeker's employment intention or directly modifies the job seeker's employment boundaries is entered into semantic segmentation. Job descriptions, explanations from service personnel, and irrelevant small talk only serve as contextual boundaries in segment localization.

[0028] The data processing server references job seekers' expressions of employment intentions and the sequence of their interactive text, dividing the text into semantic segments based on semantic integrity. A semantic segment is the smallest processing unit in the interactive text capable of independently expressing a meaning related to employment matching. Its division boundary is jointly determined by sentence boundaries, adversative conjunctions, conditional conjunctions, concessive conjunctions, exception prompts, and subsequent correction prompts. The data processing server first forms initial segments based on sentence boundaries, then performs secondary segmentation based on the conditional expression boundaries. During secondary segmentation, a dominance retention rule is applied to conditional conjunctions, concessive conjunctions, and exception prompts. The premise governed by a conditional conjunction and its corresponding intention predicate belong to the same semantic segment; when the concessive conjunction points to different constraint objects, it is split into different semantic segments according to the constraint object; when the concessive conjunction points to the same constraint object and the latter segment supplements or narrows the former, its source position order is preserved, and comparable semantic segments are formed separately. The exception content governed by exception prompts remains related to its changed intention orientation, and the exception content enters the subsequent job search constraint candidates as a condition for its validity.

[0029] When a sentence contains multiple constraint objects, the data processing server splits it into multiple semantic segments according to the predicate dominance of the constraint objects. After splitting, each semantic segment generates a source location, which is composed of semantic segment order, character range, or interaction time order identifier, and maintains a correspondence with the original interactive text. When the source location simultaneously contains semantic segment order, character range, and interaction time order identifier, the data processing server prioritizes determining the order based on the semantic segment order; if the semantic segment orders are the same, the order is determined based on the starting position of the character range; if they still cannot be distinguished, the order is determined based on the interaction time order identifier.

[0030] For example, if a job seeker states that they are okay with day shifts but not night shifts, and that they are willing to accept a longer commute if the company provides dormitory accommodation, the data processing server will categorize "okay with day shifts" as a semantic segment related to working hours, "not okay with night shifts" as another semantic segment under the same constraint, and "if the company provides dormitory accommodation, I'm willing to accept a longer commute" as a semantic segment related to commuting distance and accommodation conditions. These three semantic segments are then assigned their source locations in the order they appear and used as the basis for determining subsequent modifications, exceptions, and coverage relationships when forming job application constraint candidates.

[0031] After semantic segmentation, the data processing server performs an employment matching relevance determination on each semantic segment. First, following the naming rules for constraint objects in the constraint expression system, the server searches for terms within the semantic segment that can be categorized as working hours, commuting distance, accommodation conditions, salary arrangements, physical endurance, and family care arrangements. Then, it determines whether the term is in the same semantic segment or the same dominance chain as a willingness trigger term for acceptance, exclusion, or conditional acceptance. If both the term and the willingness trigger term are true, the semantic segment is determined to be employment-matching. Segments that do not form employment matching constraints are not included in the job search constraint candidate set; segments that do form employment matching constraints, along with their source locations, are output as a sequence of semantic segments to be identified.

[0032] The data processing server takes a sequence of semantic segments to be identified as input and performs conditional expression recognition on each segment. For each semantic segment, the server generates a sequence of dependency relations consisting of word order, trigger word categories, and governing objects. Trigger word categories include conditional trigger words, concessional trigger words, exception trigger words, negation trigger words, and willingness trigger words. Conditional trigger words include if, as long as, premise is, have, provide, satisfy; concessional trigger words include but, however, also, still acceptable; exception trigger words include unless, only, limited to, temporary; negation trigger words include no, cannot, do not accept, do not consider; willingness trigger words include can, accept, consider, willing.

[0033] The data processing server first locates the intention trigger word, then searches forward for the nearest conditional trigger word, concession trigger word, or exception trigger word that does not cross the semantic segment boundary. The noun or predicate content governed by this trigger word is written as a condition for acceptance. If a negative trigger word precedes the intention trigger word and no conditional, concessional, or exception trigger word appears between them, the intention orientation is written as exclusion. If the intention trigger word shares the same governing chain as the conditional, concessional, or exception trigger word, the intention orientation is written as conditional acceptance. Other affirmative intention expressions are written as acceptance. Based on the naming rules for constraint objects in the constraint expression system, the data processing server merges different natural language expressions into a unified constraint object, enabling subsequent modules to perform grouped processing under the same constraint object.

[0034] In practical implementation, for semantic fragments where accommodation is provided and a commute exceeding one hour is acceptable, the data processing server identifies the commute distance as a constraint, "can also" as conditionally acceptable, and "providing accommodation" as a condition for acceptance. It also records the order of appearance of the semantic fragment based on its source location. For semantic fragments where night shifts are not acceptable unless it's just a temporary cover shift, the data processing server identifies the working hours as a constraint. It first identifies night shifts as unacceptable and excludes them, then identifies "unless it's just a temporary cover shift" as an exception to this exclusion. This generates a conditionally acceptable expression with a condition for acceptance, enabling subsequent modules to determine whether the candidate option conflicts with other candidates under the same constraint due to contextual conflict, order correction, or regular merging.

[0035] When a semantic fragment contains multiple conditional premises, the data processing server determines the attribution relationship based on the dominance distance between the conditional premises and the intention predicate. When multiple premises jointly govern the same intention predicate, the fulfillment conditions are written into the same candidate option using a conjunction relation. A conjunction relation indicates that multiple fulfillment conditions must be met simultaneously. In subsequent job matching determination, the data processing server considers the conjunction relation satisfied only when the fulfillment conditions already possessed by the job in the job performance feature set cover all the fulfillment conditions in the conjunction relation. If any fulfillment condition is in an unsatisfied state, the conjunction relation is not satisfied. If any fulfillment condition is in an ambiguous state and there are no unsatisfied states, the corresponding condition of the conjunction relation is satisfied. When different intention predicates are governed by different premises, they are split into different job application constraint candidates, each inheriting its respective source position. Thus, the data processing server outputs a sequence of constraint content with determined constraint objects, intention orientations, fulfillment conditions, and source positions.

[0036] The data processing server references the sequence of constraint content, converting each constraint into a candidate job-seeking constraint, and then aggregates them to form a candidate set of job-seeking constraints. The conversion process follows a unified writing order: first, the constraint object is written; then, the intention orientation is written; then, the conditions for acceptance are written; and finally, the source location is written. The source location directly inherits the semantic fragment order, character range, or interaction time order identifier generated during semantic fragment segmentation, without reordering. For acceptance or exclusion without explicit preconditions, the conditions for acceptance are written as empty conditions; for expressions with preconditions, concessions, or exceptions, the conditions for acceptance are written with the corresponding precondition content, maintaining the association with the source location.

[0037] After the job search constraint candidate set is generated, the data processing server writes the set state to the readable state. When the set state is readable, the constraint graphing module reads the job search constraint candidate set and performs constraint relationship reconstruction. When the set state is missing, the data processing server rereads the semantic fragment segmentation results and conditional expression recognition results, fills in the constraint object, intention orientation, establishment conditions and source location before entering the constraint graphing module.

[0038] The constraint graphing module receives a set of job application constraint candidates, groups the job application constraint candidates by constraint object, and performs constraint synthesis and discrimination based on the compatibility of the conditions, the opposite of the intention orientation, and the order of the source position to generate a job application constraint graph.

[0039] The constraint mapping module starts with the job constraint candidate set and its readable status output by the intention extraction module. The data processing server first reads the constraint object, intention orientation, establishment conditions, and source location of each job constraint candidate. When the set status indicates that the job constraint candidate set has completed the identification of the expression subject, semantic segmentation, and conditional expression identification, it enters the constraint relationship reconstruction process.

[0040] The data processing server groups job application constraint candidates according to the constraint objects. Before grouping, the server first reads the constraint objects of each candidate job application constraint according to the naming rules of the constraint representation system. Job application constraint candidates with the same constraint object normalization result are grouped into the same candidate group. Candidate Group For the first Each candidate group, derived from the job-seeking constraint candidate set, serves to limit the comparison scope for contextual conflict and order correction. After the candidate groups are formed, the data processing server sorts the job-seeking constraint candidates within each group based on their source location, creating a constraint expression sequence. When the source location contains semantic fragment order, character range, or interaction time order identifiers, sorting is prioritized according to semantic fragment order; if semantic fragment orders are the same, they are then sorted according to the starting position of the character range; if still indistinguishable, they are sorted according to interaction time order.

[0041] In practice, when the job application constraint candidate set includes options regarding working hours such as "day shift is acceptable," "night shift is not acceptable," and "night shift is acceptable if it's just a temporary cover," the data processing server categorizes these options into the candidate group corresponding to the working hours and forms a constraint expression sequence based on their source location, pointing from "day shift is acceptable" to "night shift is not acceptable," and then to "night shift is acceptable if it's just a temporary cover." This constraint expression sequence does not change the original content of the job application constraint candidate options; it only determines the comparison order of the expressions under the same constraint object.

[0042] The data processing server references candidate groups and constraint expression sequences, performs pairwise comparisons on job application constraint candidates under the same constraint object, and calculates contextual conflicts. For the first... Candidate groups The data processing server will form a candidate pairing sequence from any two different job application constraints within the candidate group. The same candidate cannot be paired with itself, and two candidates that have already been compared cannot be paired again. For the first Candidate pairing sequences in each candidate group, derived from candidate groups The job-seeking constraint candidates within the context serve to provide pairwise comparison objects for calculating contextual conflict.

[0043] Each candidate pairing first undergoes a conditional relationship determination, followed by a preference relationship determination. Conditional relationships include identical, overlapping, compatible, and mutually exclusive. Two conditions are considered identical if their unified condition objects, condition attributes, and condition values ​​are all identical. If one condition is an empty condition and the other is a specific premise, the empty condition covers the specific premise. Two conditions are considered overlapping if their condition objects and condition attributes are identical, and one condition's value range includes the other's value range. When the condition value is numerical, the data processing server first converts the corresponding unit to a pre-defined unified unit in the constraint expression system, then compares the inclusion relationships of numerical intervals, upper limits, or lower limits. When the condition value is enumerable, the data processing server determines the inclusion relationship based on the pre-defined enumeristic hierarchy in the constraint expression system. Enumeristic hierarchy relationships include providing accommodation covering providing employee dormitories in accommodation conditions, and night shifts covering temporary shifts in working hours. When the condition value is factual, the data processing server determines whether it can cover based on the state relationships between "possible," "not possessed," and "unclear." Two conditions are considered compatible if they can both be true and there are no opposite values. Two conditions are considered mutually exclusive if they have incompatible values ​​under the same condition object and attribute. Mutually exclusive relationships are determined by the condition value relationships in the constraint expression system, including opposite states of possession and non-possession under the same condition object and attribute, range states where two numerical intervals do not intersect, and mutually exclusive enumeration relationships pre-defined by employment service business rules. These include mutual exclusion of providing accommodation under accommodation conditions and not providing accommodation under working hours conditions, and mutual exclusion of fixed night shifts and not accepting night shifts under working hours. Candidate pairs with identical, overlapping, or compatible condition relationships are counted in the number of comparable candidate pair pairs. In comparable candidate pairings, if the intended orientations are acceptance and exclusion, or conditional acceptance and exclusion, or acceptance and exclusion under the same conditions, these are counted in the number of conflict records. .

[0044] For the Candidate groups Conflict within the same territory We obtain it according to the following formula: ,in, For the first The contextual conflict of each candidate group is measured on a dimensionless scale and is used to determine whether there are opposing intentions in the same or compatible conditional context under the same constraint object. The number of conflict records in the same context is derived from pairings in the candidate pairing sequence that simultaneously meet the conditions of being comparable and having opposite intentions. The number of comparable candidate pairings is derived from candidate pairings whose conditions for establishment are identical, overlapping, or compatible. When the value is 0, there is conflict within the same territory. Write 0 and proceed to the regular merge candidate path.

[0045] In this embodiment, among the candidate groups regarding commuting distance, there are options for commuting within 1 hour and options for commuting further if accommodation is provided. These two options do not have opposite intentions, and the data processing server does not generate a context conflict record. If there is also an option in the same candidate group that does not accept commuting beyond 1 hour, then a comparison needs to be made between this option and the option that accepts commuting further if accommodation is provided. When providing accommodation can explain the premise of accepting commuting further, and the condition of not accepting commuting beyond 1 hour is empty, the two have a conditional compatibility basis and opposite intentions, and a context conflict record is generated.

[0046] The data processing server references the constraint expression sequence and contextual conflict records to identify, within the same candidate group, whether job constraint candidates from later source positions supplement, narrow, or modify job constraint candidates from earlier source positions, and calculates the order correction. The data processing server reads candidate groups in ascending order of source position. Job search constraints are selected from the candidate options, and a sequential pairing sequence is formed. Sequential pairing consists only of first-source position candidates and second-source position candidates, and both must belong to the same candidate group. For the first Within each candidate group, a sequence of paired sequences formed according to their source location originates from the candidate group. Its function is to provide comparison objects for order correction calculation.

[0047] The data processing server determines the order comparison range based on the source location. When the source location includes a semantic fragment order, candidates from later source locations are considered adjacent if the semantic fragment order difference between them and candidates from earlier source locations is 1; candidates from earlier source locations are considered close neighbors if the semantic fragment order difference is greater than 1 but not greater than a preset number of nearest neighbor fragments. The preset number of nearest neighbor fragments is configured by the employment service business rules before implementation, and in one implementation, it is set to 2 to cover situations where job seekers interpret the previous expression in the same round of answers or immediately following supplementary answers. When the source location includes an interaction time order, candidates from later source locations are considered close neighbors if they are in the same round of interaction or the next round of interaction. Only adjacent or near neighbor candidates are included in the number of candidate pairings that can form an order comparison. .

[0048] Enter In sequential pairing, if the willingness of a later-source position candidate changes relative to the earlier-source position candidate (acceptance, exclusion, or conditional acceptance), and the conditions for the establishment of the later-source position candidate can more specifically limit, supplement, or narrow the scope of application of the earlier-source position candidate, this change is included in the number of records for which the adjustment has been established. When the candidate for the first source position is empty, and the candidate for the second source position has a specific condition that can explain the applicable boundary of the first expression, a narrowing relationship is identified. When both candidates have conditions, the data processing server compares their scope according to the condition normalization rule in the constraint expression system. If the condition range of the second source position candidate is less than or equal to the condition range of the first source position candidate, a narrowing relationship is identified.

[0049] For the Candidate groups , order correction We obtain it according to the following formula: ,in, For the first The ordinal correction of each candidate group is a dimensionless scale used to determine whether the expression at the later source position complements, narrows, or modifies the expression at the earlier source position. To correct the number of established records, the results are derived from sequential pairings that satisfy the relationships of subsequent source location, changes in intention orientation, and narrowing of establishment conditions; The number of candidate pairings that can enter the sequential comparison range is derived from adjacent or nearest neighbor candidates. When When it is 0, the sequence is corrected. Write 0.

[0050] In this embodiment, if a job seeker initially states that night shifts are unacceptable, but then adds that they would accept temporary night shift work, the data processing server reads the candidate from the constraint expression sequence corresponding to the working hours. The job seeker's preference changes from exclusion to conditional acceptance. The condition that the night shift is unacceptable only as a temporary work stoppage provides an exception to the existing constraint, thus generating a corrected record, which is used as the basis for order correction calculations. If a job seeker initially states that a long commute is unacceptable, but then states after several rounds that they desire a higher salary, the candidate from the subsequent source location does not point to the same constraint object and is not compared within the same candidate group.

[0051] The data processing server organizes the contextual conflict, ordinal correction, contextual conflict record, and correction validity record of each candidate group into constraint synthesis input, and then passes the constraint synthesis input to the hierarchical evidence reasoning model. The hierarchical evidence reasoning model adopts two-level rule reasoning. The first level is the conflict existence judgment, where the data processing server determines the contextual conflict... Conflict determination threshold Compare; when When the evidence of conflict is not established, the output of the composite discrimination result is a standard merge; when At this point, the evidence of a conflict is established, and the process proceeds to the second level of judgment. The second level involves subsequent corrective judgment, where the data processing server corrects the sequence number. With the correction judgment threshold Compare; when When subsequent corrected evidence is valid, the output of the synthetic discrimination result covers and eliminates the problem; when If subsequent corrective evidence is found to be invalid, the composite discrimination results are output and retained in parallel.

[0052] Conflict determination threshold Calculated from candidate groups whose existence of genuine conflicts has been manually verified. The data processing server calculates intra-regional conflicts for each of these candidate groups. Sort the obtained values ​​in ascending order, for example, take the upper quartile as... Correct the judgment threshold. The order corrections are calculated from candidate groups whose existence of subsequent correction relationships has been manually confirmed. The data processing server calculates the order corrections for each of these candidate groups. Sort the obtained values ​​in ascending order, for example, take the upper quartile as... When the implementation scenario has not yet accumulated manual confirmation records, the data processing server uses an initial judgment table configured by the employment service business rules. This initial judgment table at least specifies the judgment boundaries for when a conflict within the same context meets the conditions for conflict existence and when order correction meets the conditions for subsequent correction. The initial judgment table uses candidate groups as the judgment objects and includes at least the same-context conflict interval, the order correction interval, evidence of conflict existence, evidence of subsequent correction, and a composite judgment result. When the data processing server reads the initial judgment table, it first determines the evidence of conflict existence based on the interval to which the same-context conflict belongs, then determines the evidence of subsequent correction based on the interval to which the order correction belongs, and finally outputs a composite judgment result from one of the following: conventional merging, overwrite resolution, or parallel retention.

[0053] The data processing server uses the output of the hierarchical evidence reasoning model as the synthetic discrimination result and intervenes in the generation of edge attributes in the job application condition constraint graph based on the synthetic discrimination result. For candidate groups whose synthetic discrimination result is a regular merging, the data processing server merges job application constraint candidates with semantic consistency or complementary conditions into the same constraint expression, generates master nodes based on constraint objects, generates condition nodes based on the conditions for fulfillment, and generates edge attributes connecting master nodes and condition nodes based on intention orientation. For candidate groups whose synthetic discrimination result is a covering resolution, the data processing server uses job application constraint candidates corresponding to the later source position to participate in edge attribute generation. Covered candidates are not used as valid constraint edges in job matching determination, but their source position, constraint object, and covered relationship are jointly saved as constraint basis along with the candidates from the later source position. For candidate groups whose synthetic discrimination result is a parallel retention, the data processing server writes job application constraint candidates that conflict with each other but whose correction relationship cannot be confirmed into the job application condition constraint graph, and retains multiple parallel constraint edges under the same constraint object master node.

[0054] After the job application constraint graph is generated, the data processing server saves the corresponding constraint object, fulfillment condition, willingness orientation, source location, and composite judgment result for each constraint edge. The job application constraint graph uses constraint objects as master nodes, fulfillment conditions as condition nodes, and willingness orientation as the edge attribute connecting master nodes and condition nodes, allowing multiple parallel constraint edges under the same constraint object. This job application constraint graph serves as input to the job determination module, used for matching and determining the job performance feature set of the employment opportunities to be matched.

[0055] The job determination module obtains job information for potential employment opportunities and matches them with job seeker requirements. Figure 1 The constraint expression system is subjected to condition normalization and performance feature extraction to generate a job constraint judgment table.

[0056] The job determination module starts with the job requirement constraint graph output by the constraint graph module and the job information of the employment opportunities to be matched. The data processing server first reads the main node of the constraint object in the job requirement constraint graph and uses this main node as the target range for job information parsing; then it obtains the job information of the employment opportunities to be matched, which includes the job posting text, job field records, and supplementary descriptions from the recruiter regarding working hours, commuting arrangements, accommodation, salary arrangements, physical requirements, and family care convenience.

[0057] The data processing server aligns job information according to the naming rules of constraint objects in the constraint representation system. During processing, job information is first divided into job description fragments, and then the job items corresponding to the main nodes in the job application condition constraint diagram are extracted from each job description fragment. When the natural language expression in the job description fragment is inconsistent with the main node name, it is grouped into the same constraint object according to the synonym representation mapping rules. For example, if the job information includes "accommodation provided," "employee dormitory provided," or "factory dormitory available for application," it is aligned to the accommodation conditions; if it includes "two shifts," "night shifts," or "occasional night shifts," it is aligned to the working hours.

[0058] In practice, if the job application constraint graph includes three main nodes: working hours, commuting distance, and accommodation, the data processing server reads descriptions from a job posting such as "working hours are from 8:00 AM to 5:00 PM," "dormitory provided," and "approximately 12 kilometers from the job seeker's residence." It then aligns "8:00 AM to 5:00 PM" to working hours, "dormitory provided" to accommodation, and "approximately 12 kilometers" to commuting distance. After alignment, the job information forms a sequence of job description fragments organized by constraint objects and is then passed to the condition normalization and performance feature extraction stages.

[0059] The data processing server references a sequence of job description fragments and uses the condition nodes and constraint edges in the job application constraint graph as normalization references. The server first reads the constraint objects, fulfillment conditions, and willingness orientations connected by each constraint edge, and then extracts job performance content from the aligned job description fragments that can prove whether the job meets the corresponding conditions. Job performance content consists of factual descriptions or field values ​​in the job information that can be directly compared with the fulfillment conditions.

[0060] Condition normalization is performed according to the condition normalization rules in the constraint expression system. The data processing server converts the job performance content into job application condition constraints. Figure 1 The conditional expression is transformed in the following order: first, determine the constraint objects corresponding to the job performance content; second, determine whether they can cover the conditions in the condition nodes; and finally, write the verifiable job conditions into the job performance feature set. For numerical or range-type conditions, the data processing server compares them according to the inclusion relationship between the job field values ​​and the conditions; for factual conditions, it compares them according to whether the job information explicitly provides the corresponding facts; for conditions that do not appear in the job information, the unspecified performance conditions of the job are written.

[0061] When a job posting explicitly states a certain condition, that condition is written into the job's existing performance conditions. When a job posting explicitly states a fact contrary to the condition, that condition is written into the job's unmet performance conditions. When a job posting only states facts that require further confirmation from the recruiter, such as "applicable," "negotiable," "discussable in person," "to be confirmed separately," or "determined based on scheduling," that condition is written into the job's undefined performance conditions. When a job posting completely lacks a corresponding constraint description, the data processing server writes that constraint into the job's undefined performance conditions. This undefined state serves only as an intermediate state within the job's performance feature set. When it enters the job constraint determination table, the data processing server further transforms it into a satisfied state, a condition satisfied state, or a unsatisfied state based on the willingness, condition, and existence of a verifiable statement of the corresponding valid constraint edge.

[0062] In practice, if the job application constraint graph contains the constraint edge "Commuting Distance - Accommodation Provided - Conditional Acceptance," the data processing server reads the provision of employee dormitory information from the job posting, normalizes the job's performance content to "Accommodation Provided," and writes it into the job performance feature set as a performance condition already possessed by the job. If the job posting only states that the distance is far without specifying accommodation arrangements, the job performance feature item corresponding to commuting distance is written into the unspecified performance condition of the job. Similarly, if the job application constraint graph contains the constraint edge "Working Hours - Temporary Substitute - Conditional Acceptance," and the job posting states that temporary substitute work is required 1 to 2 times per month, the data processing server normalizes this description to a temporary substitute work condition and includes it as a performance condition already possessed by the job corresponding to the working hours.

[0063] The job performance feature set is stored as a normalized result, organized by job opportunity. Each job opportunity corresponds to a job performance feature set. Within the same job performance feature set, the set stores the fulfilled performance conditions, unmet performance conditions, and unspecified performance conditions of the job, categorized by constraint object. After this set is generated, the data processing server establishes a matching relationship between it and the job application condition constraint graph, and performs constraint satisfaction determination.

[0064] The data processing server takes the job performance feature set and job application condition constraint graph as input, and performs graph structure traversal matching for each job opportunity to be matched. Valid constraint edges are those retained in the job application condition constraint graph after synthesis and participation in the job performance feature set matching determination. Covered candidates corresponding to coverage resolution are not considered valid constraint edges; multiple constraint edges corresponding to parallel retention are all considered valid constraint edges; and merged constraint edges corresponding to regular merging are considered valid constraint edges.

[0065] The data processing server first reads the constraint objects in the order of the main nodes of the job application condition constraint graph, and then locates the condition nodes and valid constraint edges under the constraint object. Then, it searches for the job performance feature items corresponding to the same constraint object in the job performance feature set, and compares the performance conditions that the job has, the performance conditions that the job has not met, and the performance conditions that the job has not specified with the condition nodes one by one.

[0066] For constraint edges with an acceptance orientation, if the job performance feature set has a corresponding constraint object and there are no unmet performance conditions for the job under that constraint object, write the "satisfied" state; if the job performance feature set has a corresponding constraint object but its necessary job performance content is unclear, and the job information contains confirmable statements such as "applyable," "negotiable," "discussable," "to be confirmed separately," or "determined according to the schedule," write the "condition satisfied" state; if the job performance feature set explicitly states that the job does not have the corresponding necessary job performance content, or if the job information completely lacks the constraint object and there are no confirmable statements, write the "dissatisfied" state. For constraint edges with an exclusion orientation, if the job performance feature set explicitly includes the excluded content, write the "dissatisfied" state; if it explicitly does not include the excluded content or the job description is inconsistent with the excluded content, write the "satisfied" state; if the job information is missing and it cannot be determined whether it includes the excluded content, write the "condition satisfied" state and associate it with the job performance content that needs to be confirmed. For constraint edges with a willingness orientation of conditional acceptance, the state is written as satisfied when the performance conditions already met by the position cover the establishment conditions; the state is written as satisfied when the performance conditions not explicitly met by the position correspond to the establishment conditions; and the state is written as unsatisfied when the performance conditions not met by the position correspond to the establishment conditions.

[0067] When multiple parallel constraint edges exist under the same constraint object in the job application constraint graph, the data processing server reads the edge attributes according to the composite judgment results retained by the constraint graph construction module. For constraint edges formed by regular merging, a judgment is performed based on the merged constraint expression; for constraint edges formed by overlay resolution, only valid constraint edges corresponding to the later source positions participate in the judgment; for multiple constraint edges formed by parallel retention, judgments are performed separately, and their respective establishment conditions and intention orientations are written into different judgment results under the same constraint object.

[0068] In practice, under the main node of working hours in a job application constraint graph, constraints excluding night shifts are retained. If the job performance feature set shows that the job involves two shifts and includes night shifts, the data processing server writes the unmet status under the working hours constraint object for this job opportunity. If another constraint edge states that night shifts are acceptable if it is only a temporary cover, and the job performance feature set shows that night shifts are only temporary cover, then this constraint edge is written as satisfied. For conditional acceptance edges under the commuting distance main node, if the condition is that accommodation is provided, and the job performance feature set has already written that accommodation is provided, then the job opportunity is written as satisfied under the commuting distance constraint object. If the job information indicates that accommodation can be applied for but does not explicitly state the result, then the condition is written as satisfied, and the corresponding condition is retained for the recommendation control module to generate condition prompts.

[0069] The data processing server references the judgment results of each constraint object, summarizes them according to the job opportunities to be matched, and generates a job constraint judgment table. The data processing server treats the job opportunities to be matched as row-level processing objects, and uses the constraint objects and valid constraint edges in the job application condition constraint graph as the judgment criteria. It writes the satisfied state, condition satisfied state, and unsatisfied state into the corresponding positions, and retains the constraint object that caused the state, the conditions for its establishment, and the intention orientation. For records with a satisfied state, the data processing server retains the satisfied constraint objects and corresponding performance conditions; for records with a condition satisfied state, it retains the conditions that still need to be established or confirmed; for records with an unsatisfied state, it retains the constraint edge that triggered the unsatisfied state and its intention orientation, enabling the recommendation control module to perform filtering, condition display, or sorting control accordingly.

[0070] The recommendation control module filters, displays, and sorts the job opportunities to be matched based on the job constraint judgment table, and outputs job recommendation results that include the recommendation order, constraint basis, and condition prompts.

[0071] The recommendation control module uses the job constraint judgment table output by the job judgment module as its starting point and simultaneously reads the job performance feature set generated by the job judgment module and the job application condition constraint graph generated by the constraint graph construction module. The data processing server first reads the judgment status in the job constraint judgment table item by item according to the job opportunities to be matched, and then reads the constraint object, the conditions for its establishment, and the intention orientation that generated the judgment status, so that each job opportunity to be matched can be mapped to the main node, condition node, and effective constraint edge in the job application condition constraint graph.

[0072] The data processing server executes recommended control routing in the order of processing unmet states, satisfied states, and satisfied states. For any job opportunity to be matched, if there is an unmet state in the job constraint judgment table, and this unmet state is caused by an exclusionary constraint conflict or the failure to meet the necessary conditions, the data processing server writes the job opportunity to be matched into the filtering candidate queue; for job opportunities to be matched that do not have unmet states but have satisfied states, they are written into the condition display candidate queue; for job opportunities to be matched where all valid constraint edges are satisfied, they are written into the sorting candidate queue. All of the above queues are based on the job constraint judgment table and do not change the job performance feature set and job application condition constraint graph.

[0073] In practice, if a job posting shows "unmet" working hours in the job constraint judgment table, it's because the job performance feature set indicates the job includes night shifts, while the corresponding constraint edge in the job application condition constraint graph indicates exclusion. The data processing server then adds this job posting to the filter candidate queue. If another job posting is "met" under the commuting distance constraint, it's because the condition for fulfillment is providing accommodation, while the job information only indicates accommodation is available. The data processing server adds this job posting to the condition display candidate queue, retaining the condition for providing accommodation. If a job is satisfied in terms of working hours, commuting distance, accommodation, and compensation, it enters the sorting candidate queue.

[0074] The data processing server references a filtering candidate queue and a conditional display candidate queue, and performs recommendation control accordingly. For job opportunities to be matched in the filtering candidate queue, the data processing server reads the constraint objects, conditions for their establishment, and willingness orientations that trigger the unmet status, and removes the job opportunity from the recommended output range; at the same time, it retains the corresponding constraint basis, allowing employment service personnel to view the reason why the position was filtered. If the unmet status stems from a conflict between the job performance feature set and exclusionary constraints, the data processing server writes the constraint object and corresponding willingness orientation into the filtering basis; if the unmet status stems from the conditional acceptance not being covered by the job performance feature set, the data processing server writes the unmet condition into the filtering basis; if the same job opportunity has multiple unmet statuses, the filtering basis is saved sequentially according to the order of the constraint objects in the job constraint judgment table. Filtered job opportunities are not displayed to job seekers, and the constraint objects, conditions for their establishment, and willingness orientations that trigger the unmet status are saved as constraint basis for employment service personnel to verify the reason for recommendation control.

[0075] For job opportunities awaiting matching in the candidate queue, the data processing server retains their condition display status and associates the conditions for achieving a "condition met" status in the job constraint judgment table with the corresponding constraint objects. This association path involves first reading the constraint object containing the "condition met" status, then locating the corresponding condition node in the job application constraint diagram, followed by reading the job conditions that still need confirmation or are yet to be met from the job performance feature set, and finally generating condition prompts. Job information is only written to the "condition met" status when it contains job performance content indicating that the recruiter can still confirm or arrange it, such as "applyable," "negotiable," "negotiable," "to be confirmed separately," or "determined according to schedule." If the job information does not contain any corresponding performance content, or if the job information explicitly records facts contrary to the conditions, it is written to the "condition not met" status. Condition prompts are derived solely from the conditions in the job constraint judgment table and the job application constraint diagram, without introducing new job application boundaries.

[0076] In this embodiment, a certain position has a long commute, but the recruiter states that employee dormitory accommodation is available. The corresponding constraint edge in the job application constraint graph indicates conditional acceptance upon accommodation provision. The job constraint decision table records this position as condition-met. The data processing server retains the condition display status of this position and adds "accommodation needs confirmation" to the condition prompt. If a position meets both the working hours and salary arrangements, but only the accommodation condition is pending confirmation, then this position is not filtered but instead appears in the job recommendation results as condition-displayed.

[0077] The data processing server references the sorting candidate queue and the condition display candidate queue, and reads the job performance feature set and job application condition constraint graph corresponding to each job opportunity to be matched, and calculates the degree of satisfaction. The valid constraint edges participating in the sorting come from the constraint edges in the job application condition constraint graph that are retained by the synthesis and discrimination results of the constraint graphing module and participate in the matching judgment of the job judgment module; for constraint objects formed by overlay resolution, only the valid constraint edges corresponding to the later source positions participate in the sorting; for constraint objects formed by parallel retention, each parallel constraint edge participates in the sorting separately; for constraint objects formed by regular merging, the sorting is based on the merged constraint edges.

[0078] When calculating the degree of satisfaction, the data processing server includes both the satisfied state and the condition satisfied state in the state count. For the... One unmatched job opportunity, degree of satisfaction We obtain it according to the following formula: ; in, For the first The degree to which each job opportunity is satisfied is used to generate the recommendation order; For the first The number of valid constraint edges for each unmatched job opportunity to achieve a satisfied state is derived from the job constraint judgment table. For the first The number of valid constraint edges for each unmatched job opportunity meeting the conditions is derived from the job constraint judgment table. For the first The total number of valid constraint edges used in the determination of each unmatched job opportunity is derived from the matching range between the job application condition constraint graph and the job performance feature set; a coefficient is included for condition satisfaction, used to indicate that the condition satisfaction state is lower than the satisfied state but higher than the unsatisfied state. The value is configured by the employment service business rules before implementation, and is greater than 0 and less than 1; in one implementation, it is taken as 0.5. When the employment service business rules are adjusted, the data processing server updates the conditions for meeting the inclusion coefficient according to the adjusted business rules. When no business rules are adjusted, The process remains unchanged within the same job recommendation cycle. Unsatisfactory job opportunities are added to the filtering candidate queue and are not included in the satisfaction ranking.

[0079] The data processing server generates recommendations in descending order of satisfaction level. When satisfaction levels are the same, the server arranges them in ascending order of the number of fulfilled conditions; if still the same, it arranges them in descending order of the number of fulfillment conditions already met by the job in the job performance feature set; if still the same, it arranges them stably according to the order in which the job opportunities entered the job information set. The order in which job opportunities enter the job information set is derived from the record order when the job information is read by the data processing server; when the job information comes from a database, the entry time of the job records or the retrieval return order is used; when the job information comes from a batch-imported file, the record arrangement order in the file is used. Job opportunities in the sorted candidate queue are written to the recommendation output first; job opportunities in the condition display candidate queue are written to the recommendation output according to their satisfaction level while retaining the condition display status.

[0080] In this embodiment, if job A is satisfied in all four valid constraints—working hours, commuting distance, accommodation, and salary—job A enters the ranking candidate queue and receives a higher recommendation order. If job B is satisfied in terms of working hours and salary, and is satisfied in terms of commuting distance due to the availability of accommodation, job B enters the condition display candidate queue with a condition prompt that accommodation needs to be confirmed. If job C is unsatisfied due to the inclusion of excluded night shift arrangements, job C enters the filtering candidate queue and is not included in the satisfaction ranking.

[0081] The data processing server takes the condition display status, recommendation order, and constraint basis from the job constraint judgment table as input to generate job recommendation results. The job recommendation results include at least the recommendation order, constraint basis, and condition hints. The recommendation order is derived from the satisfaction level ranking results; the constraint basis is derived from the constraint objects, conditions for fulfillment, and preferences corresponding to each judgment status in the job constraint judgment table; the condition hints are derived from the conditions that still need to be fulfilled or confirmed in the condition display candidate queue. For job opportunities in the satisfied state, the condition hints are empty or retain a "satisfied" statement; for job opportunities in the satisfied state, the condition hints include the specific conditions for fulfillment; for filtered job opportunities, the constraint basis includes the constraint objects and constraint edges that trigger the unsatisfied state, but the job opportunity will not be included in the job seeker recommendation display scope.

[0082] The data processing server transforms job seekers' employment service interaction texts into a candidate set of job search constraints via the intention extraction module and generates a job search condition constraint graph through contextual conflict, ordinal correction, and hierarchical evidence reasoning models. Subsequently, the job information of the job opportunities to be matched is transformed into a set of job performance features and matched with the job search condition constraint graph to form a job constraint judgment table. Finally, filtering, condition display, and sorting are performed based on the job constraint judgment table to output job recommendation results that include recommendation order, constraint basis, and condition prompts.

[0083] The above embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An AI-driven employment supply and demand precise matching system for a key group, characterized in that, It includes a willingness extraction module, a constraint graph construction module, a job determination module, and a recommendation control module; The intention extraction module obtains interactive text that represents job seekers' employment intentions, performs subject identification, semantic segmentation, and conditional expression identification on the interactive text, and generates a candidate set of job application constraints that includes the constraint object, intention orientation, conditions for establishment, and source location. The constraint graphing module receives a set of job application constraint candidates, groups the set of job application constraint candidates according to the constraint objects, and performs constraint synthesis and discrimination based on the compatibility of the conditions, the opposite relationship of the intention orientation, and the order of the source location to generate a job application constraint graph. The job determination module obtains job information for job opportunities to be matched, performs condition normalization and performance feature extraction according to the constraint expression system consistent with the job application condition constraint diagram, and generates a job constraint determination table. The recommendation control module filters, displays, and sorts the job opportunities to be matched based on the job constraint judgment table, and outputs job recommendation results including recommendation order, constraint basis, and condition prompts.

2. The AI-driven employment supply and demand precise matching system for a specific group according to claim 1, characterized in that, The data processing server performs basic text normalization on the interactive text, processing sentence break markers, pause markers, and transcription noise markers into recognizable sentence boundaries, and forming an interactive text sequence according to the time of interaction. The subject of expression identification determines the job seeker's expression of employment intention based on speaker identifiers, title references, employment intention predicates, and job description predicates. The rounds and character ranges of non-job seeker employment intention expressions are retained as source location references, and the text belonging to the job seeker's employment intention expression is sent to semantic segmentation.

3. The AI-driven employment supply and demand precise matching system for a specific group according to claim 2, characterized in that, Semantic segmentation begins with sentence boundaries to form initial segments, and then proceeds to secondary segmentation based on the dominance relationships of conditional conjunctions, concessive conjunctions, and exception markers. The data processing server forms a dependency sequence based on condition trigger words, concession trigger words, exception trigger words, negation trigger words, and willingness trigger words. The dependency sequence determines the constraint object, willingness orientation, conditions for establishment, and source location. The constraint content of each semantic fragment is written into the job application constraint candidate set in a unified writing order.

4. The AI-driven employment supply and demand precise matching system for a specific group according to claim 1, characterized in that, The job constraint candidate set is grouped according to the naming rules of constraint objects in the constraint expression system. Job constraint candidates with the same constraint object normalization result are grouped into the same candidate group. The data processing server sorts the job constraint candidates in the candidate group according to the source location to form a constraint expression sequence. The item words that do not match the constraint object naming rules are written into the constraint objects to be confirmed. The confirmed constraint objects continue to participate in the grouping and constraint relationship reconstruction process.

5. The AI-driven precise matching system for employment supply and demand targeting key groups as described in claim 4, characterized in that, Constrained synthesis discrimination includes the calculation of same-context conflict. The data processing server forms a candidate pairing sequence within the same candidate group and compares the establishment conditions and intention orientation of each candidate pairing. When the conditions for establishment are identical, overlapping, or compatible, and the intention orientations are acceptance and exclusion, conditional acceptance and exclusion, or acceptance and exclusion under identical conditions, they are included in the same-situation conflict record, and a same-situation conflict is formed based on the number of such records and comparable candidate pairings.

6. The AI-driven employment supply and demand precise matching system for a specific group according to claim 5, characterized in that, The constraint synthesis discrimination also includes ordinal correction calculation and hierarchical evidence reasoning. The data processing server forms a sequential pairing sequence according to the source position, generates a corrected establishment record based on the subsequent source position, changes in intention orientation and narrowing relationship of establishment conditions, and outputs conventional merging, overlay resolution or parallel retention according to the same context conflict and ordinal correction. It also controls the generation of main nodes, condition nodes and edge attributes of job application condition constraint graph, and retains the corresponding source position as constraint basis data.

7. The AI-driven employment supply and demand precise matching system for a specific group according to claim 1, characterized in that, Job information is aligned with constraint objects through a constraint expression system, which includes constraint object naming rules, condition normalization rules, and performance attribute expression rules. The data processing server first divides the job information into job description fragments, then assigns the job items in the job description fragments to constraint objects consistent with the main nodes of the job application condition constraint diagram, and forms a sequence of job description fragments organized by constraint objects, which is used as input for condition normalization processing and performance feature extraction.

8. The AI-driven employment supply and demand precise matching system for a specific group according to claim 7, characterized in that, The performance feature extraction process converts the job performance content into a condition expression consistent with the job application condition constraint diagram and writes it into the job performance feature set. The data processing server writes the performance conditions that the job already possesses, the performance conditions that the job does not meet, and the performance conditions that the job does not specify, according to different situations: the job information clearly states the conditions that are met, the opposite facts are clearly recorded, and only the facts that need further confirmation are recorded. The corresponding constraint objects, job performance content, and condition expression results are retained.

9. The AI-driven employment supply and demand precise matching system for a specific group according to claim 8, characterized in that, The job constraint judgment table is generated based on the matching results between the job performance feature set and the job application condition constraint graph. The data processing server matches the job's existing performance conditions, unmet performance conditions, and unspecified performance conditions with the valid constraint edges in the job application condition constraint graph, and writes them into the satisfied state, condition satisfied state, and unsatisfied state according to the different intention orientations of acceptance, exclusion, and conditional acceptance. At the same time, the corresponding conditions for establishment are retained. The valid constraint edges are the constraint edges that are retained and participate in the matching after constraint synthesis and discrimination.

10. The AI-driven employment supply and demand precise matching system for a specific group according to claim 9, wherein, The data processing server performs recommendation control based on the job constraint judgment table. Job opportunities with unmet conditions are written into the filtering candidate queue, job opportunities with met conditions and no unmet conditions are written into the condition display candidate queue, and job opportunities with all valid constraint edges met are written into the sorting candidate queue. The server then generates a recommendation order based on the number of met conditions, the number of conditions met, the number of fulfillment conditions already met by the job, and the job information set.

Images