Multi-agent cooperative vocational evaluation system

Abstract

The application discloses a multi-agent cooperative occupation evaluation system and belongs to the technical field of artificial intelligence. In order to solve the technical problem that the similarity of occupation evaluation results is too high due to the incapability of processing user self-input non-standardized appeal texts in the prior art, the application provides a multi-agent cooperative occupation evaluation system. The application obtains user demographic information, a personality characteristic vector, future representation data and an appeal text, constructs a user exclusive future self-scoring agent, scores the future representation data two by two to construct a complementarity and conflict network, generates a post recommendation according to a network topology analysis result by multi-agent cooperative reasoning, and generates an individualized occupation development report in combination with user appeal. The application is mainly used for individualized occupation evaluation.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically to a career assessment system. Background Technology

[0002] In the field of career assessment, it is common practice to collect textual descriptions of users' personal career development aspirations and related target tags, followed by manual coding and pairwise scoring by career planners based on subjective experience; alternatively, a rule-based matching method based on preset keywords is used to perform shallow association calculations, thereby constructing matching logic to generate job recommendations. However, when the set of target tags provided by users is large, personnel face extremely high cognitive loads, and the scoring results are highly susceptible to interference from subjective experience.

[0003] Existing automated systems mostly rely on pre-set rule bases to match standardized tags, which cannot process non-standardized request texts entered by users themselves. This results in the inability to internalize the individual differences of users, and the similarity of career assessment results output by different users is too high, making it difficult to meet the needs of personalized recommendations. Summary of the Invention

[0004] To address the technical problem that existing technologies cannot handle user-inputted, non-standardized request texts, resulting in excessively similar career assessment results, this invention provides a multi-agent collaborative career assessment system.

[0005] This invention is achieved through the following technical solution:

[0006] A multi-agent collaborative vocational assessment system, comprising:

[0007] The information collection module is used to acquire users' demographic information, personality trait vectors, future representation sets, and user request texts.

[0008] The Dedicated Agent Construction Module is used to construct structured system prompts for a user's dedicated future self-rating agent based on demographic information and personality feature vectors, and to configure the system prompts as the inference context for the user's dedicated future self-rating agent.

[0009] The correlation scoring module is used to use the user's exclusive future self-scoring agent to score the correlation between any two representations in the future representation set, and obtain the correlation score results of each pair of representations.

[0010] The network construction module is used to construct complementary networks and conflict networks based on the relevance scoring results, respectively.

[0011] The network analysis module is used to identify hub nodes and classify community structures for complementary and conflicting networks, respectively.

[0012] The job recommendation module is used to invoke multiple intelligent agents with different functions to perform multi-agent collaborative reasoning based on personality feature vectors, future representation sets, complementary networks, and conflict networks to generate job recommendation results.

[0013] The report generation module is used to generate personalized career development reports based on job recommendation results and user request texts, following a preset template.

[0014] Furthermore, the method for constructing the complementary network is as follows:

[0015] Using future representations in the future representation set as nodes of the complementarity network, if the correlation score between two future representations is greater than or equal to a first preset threshold, an undirected edge is established between the two future representations, with the edge weight being the absolute value of the correlation score; no edge is established between representations whose correlation score is less than the first preset threshold.

[0016] Furthermore, the method for constructing the conflict network is as follows:

[0017] Using future representations in the future representation set as nodes of the complementarity network, if the correlation score between two future representations is less than or equal to a second preset threshold, an undirected edge is established between the two future representations, with the edge weight being the absolute value of the correlation score; no edge is established between representations whose correlation score is greater than the second preset threshold.

[0018] Furthermore, the specific method for identifying the Hub node is as follows:

[0019] Those selected were those with a Personality Weighted Centrality Index (PWCI) value greater than the third threshold. The nodes form a candidate Hub node set. Determine the set of candidate Hub nodes. Is it non-empty?

[0020] If the candidate Hub node set If the set of candidate Hub nodes is not empty, then the nodes in the set of candidate Hub nodes will be determined as Hub nodes.

[0021] If the candidate Hub node set If the value is empty, all nodes are sorted in descending order based on the PWCI value, and the node with the highest sorting position (Num) is selected as the Hub node.

[0022] Furthermore, the expression for the personality-weighted centrality index (PWCI) is as follows:

[0023] ;

[0024] in, As nodes in a complementary network, , A set of nodes for a complementary network; For nodes PWCI value; For nodes The set of neighboring nodes, representing the relationship with the node. There are all other future representations that are connected. ; For the edge The weights; For nodes The corresponding preset representation type; For user personality feature vectors; The personality-type compatibility function is expressed as follows:

[0025] ;

[0026] in, It is a collection of the Big Five personality traits. , As an open dimension, For the due diligence dimension, For the extroversion dimension, For the human-centered dimension, Neuroticism dimension; ,in, For open dimension Score, For the dimension of responsibility Score, For the extroversion dimension Score, For the human-friendly dimension Score, Neuroticism dimension Score; Represents the set of Big Five personality dimensions The index; For dimension The score was normalized to The standardized score after that; For dimension The global weights satisfy , express The absolute value; For dimension The standardized score baseline threshold; For dimension For preset representation types Adjustment direction, , When, it represents dimension. With representation type Positive correlation; When, it represents dimension. With preset representation type Negative correlation.

[0027] Furthermore, the community structure division of complementary and conflicting networks includes a first-stage community division and a second-stage network aggregation, specifically as follows:

[0028] Phase One Community Division:

[0029] (1) Initial partitioning

[0030] Each node in the current network to be partitioned is initialized as an independent community, such that the number of communities is the same as the number of nodes in the initial state, and each community contains only a single node; on the first execution, the current network to be partitioned is either a complementary network or a conflicting network; on subsequent executions, the current network to be partitioned is the aggregated network obtained in the second stage.

[0031] (2) Iterative optimization

[0032] Repeat the following sub-steps until the first stage terminates:

[0033] (2.1) Node traversal

[0034] In the current iteration, each node in the network is traversed in a random order;

[0035] (2.2) Candidate Community Determination

[0036] For the current node Determine the node A set of candidate communities; the set of candidate communities includes: nodes The current community and its relationship with nodes The communities to which neighboring nodes with connections belong;

[0037] (2.3) Calculation of modularity gain

[0038] For each candidate community in the candidate community set, calculate the node The increment in modularity caused by migrating to a candidate community; where the candidate community is a node. When the current community is involved, the corresponding module degree increment value is zero;

[0039] The formula for calculating the modularity gain is:

[0040] ;

[0041] in, is the modularity gain, representing the change in network modularity after a node migration; This represents the total edge weight of the network, and its value is half of the sum of the edge weights of all edges in the currently partitioned network. For nodes and nodes Edge weights between them; For nodes The weighting degree, For nodes The weighted degree, which is defined as the sum of the weights of all edges associated with the current node; For community indicator functions, when node and nodes The value is 1 if the two communities are in the same area, and 0 otherwise. For nodes Current community For nodes Current community affiliation; For nodes With nodes The coefficient of personality similarity between them;

[0042] (2.4) Optimal Community Selection

[0043] Compare the module degree increment values ​​of each candidate community, and select the candidate community with the largest module degree increment value as the target community. ;

[0044] (2.5) Node migration determination

[0045] Determine if the maximum modularity increment is greater than zero: if so, then add the node. Migrate from the current community to the target community If not, then the node Remain in the original community; no relocation will occur.

[0046] (2.6) Iteration Termination Determination

[0047] After completing the traversal of all nodes in the current round, check whether any changes in node community affiliation occurred during this round of traversal:

[0048] If the community affiliation of at least one node changes, then based on the updated community division, return to the node traversal sub-step (2.1) and continue execution;

[0049] If no node's community affiliation changes during this round, the process is considered converged, the first phase ends, and preliminary community division results are obtained.

[0050] Second-stage network aggregation:

[0051] After completing the first phase of community partitioning optimization, network aggregation is performed, which includes:

[0052] (1) Constructing supernodes

[0053] Each community identified in the first phase is abstracted as a super node, and all super nodes constitute the node set of the aggregated network.

[0054] (2) Define the edge weights between supernodes

[0055] For any two distinct supernodes and Define the edge weights between the two. for:

[0056]

[0057] in, For nodes and nodes Edge weights between them; For nodes and nodes The coefficient of personality similarity between them;

[0058] (3) Construct the aggregation network and trigger the next iteration.

[0059] Based on the set of supernodes and the edge weights between them, an undirected weighted aggregation network is constructed; the aggregation network is then used as a new input network, and the process returns to the first stage of community partitioning to continue.

[0060] Repeat the first phase of community partitioning and the second phase of network aggregation until one of the following conditions is met:

[0061] Condition 1: The results of the first phase of community delineation are exactly the same in two consecutive phases;

[0062] Condition 2: The current community structure division rounds have reached the preset maximum round threshold;

[0063] When terminating the community structure partitioning, the community structure in the currently obtained aggregated network is determined as the final community partitioning result of either a complementary network or a conflict network.

[0064] Furthermore, the expression for the personality similarity coefficient is:

[0065] ;

[0066] ;

[0067] ;

[0068] in, User personality feature vector With nodes Corresponding preset representation type The degree of compatibility, User personality feature vector With nodes Corresponding preset representation type The degree of compatibility; Preset representation type With preset representation type The degree of correlation; Preset representation type In dimensions Weight on, Preset representation type In dimensions Weights on;

[0069] Preset representation type With preset representation type correlation The expression is:

[0070] ;

[0071] ;

[0072] in, For vectors The length of the mold, For vectors The length of the module.

[0073] Furthermore, the multiple intelligent agents with different functions are specifically: a first intelligent agent, a second intelligent agent, a third intelligent agent, and a fourth intelligent agent;

[0074] The first intelligent agent issues instructions containing the task type, input data, and return data format requirements to the second, third, and fourth intelligent agents;

[0075] The second, third, and fourth agents execute instructions in parallel, each generating analysis data:

[0076] The second intelligent agent obtains personality analysis results based on personality feature vectors;

[0077] The third agent obtains representation analysis results based on future representation sets, complementary networks, and conflict networks;

[0078] The fourth agent generates job recommendation results based on personality analysis and representation analysis results; the job recommendation results include a list of recommended jobs, a matching score for each job, and the reason for the matching.

[0079] Furthermore, the fourth agent's score generates a job recommendation list and job matching degree in the following ways:

[0080] Based on the two personality dimensions with the highest scores in the user personality analysis results, job postings whose personality dimension tags cover the two highest-scoring personality dimensions are selected from a pre-set job posting database to form a candidate job posting set. For each job posting in the candidate job posting set, vector similarity index, network coverage index, and path consistency index are calculated. The vector similarity, network coverage, and path consistency indices are linearly weighted and fused to generate a job matching score between each job posting and the user. All jobs in the candidate job posting set are sorted in descending order of matching score, and the top Num2 jobs are selected to form a recommended job posting list.

[0081] Furthermore, the expression for the vector similarity is:

[0082] ;

[0083] in, For the position The required Big Five personality score threshold vector; The Big Five personality dimensions The job-specific weights represent the differences in sensitivity of different jobs to each dimension; For the current user in the dimension The standardized score is denoted as . ; For the position Requirements in dimension The scoring threshold is based on a preset value in the job database. ;

[0084] The expression for the network coverage metric is:

[0085] ;

[0086] in, For the position Complementary networks Network coverage metrics; For the position The required set of core self-representations is pre-defined based on the job database; For complementary networks The set of Hub nodes; For set With sets The cardinality of the intersection, i.e., the set of core representations of the job requirements. The set of user hub nodes The number of representations; For set The cardinality (number of elements), i.e., the job position The required total number of core representations; For set With sets Elements in the intersection; For Hub node set Elements in;

[0087] The expression for the path consistency index is:

[0088] ;

[0089] in, For the position With conflict networks Path consistency; For conflict networks In the process, from any Hub node in the Hub node set to the job position. The set of edges covered by all shortest paths in the core representation; For conflict networks In the process, from any Hub node in the Hub node set to the job position. The set of edges covered by all feasible paths representing the core representation; For the edge The weights, where , For conflict networks Nodes and edges in Belongs to set ; For conflict networks Middle The weight, and For conflict networks Nodes and edges in Belongs to set .

[0090] The beneficial effects of this invention are:

[0091] This invention replaces the manual comparison of target tags in existing technologies by constructing a user-specific future self-assessment agent. The system uses the user's background information and personality traits as system prompts for each user's agent, creating a personalized "one person, one agent" scoring mechanism. This allows the large language model to make judgments from the user's perspective. Compared to traditional rule-matching methods that can only process preset rules, this invention's agent can understand the semantic content of free text and provide differentiated scores based on the user's personality traits, improving the differentiation and accuracy of career assessment results.

[0092] Meanwhile, this invention constructs complementary and conflict networks based on the scoring results of future self-representation. It uses network topology analysis to divide core career anchor nodes and representation groups, uncovering implicit career development paths and potential conflicts. A multi-agent collaborative framework is used for reasoning analysis to obtain job recommendation results for users. Unlike existing technologies that rank jobs solely based on vector similarity, this invention uses three indicators—vector similarity, network coverage, and path consistency—to jointly determine job rankings, balancing logical rationality with personalized needs, thus solving the problems of rigid and insufficiently personalized traditional recommendation results. Attached Figure Description

[0093] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0094] Figure 1 This is a schematic diagram of the structure of a multi-agent collaborative vocational assessment system provided by an embodiment of the present invention.

[0095] Figure 2 This is a schematic diagram of a future self-complementarity network in one embodiment of the system of the present invention. The two representations connected by solid lines represent a strong complementary relationship between the two, with a corresponding complementarity score of 2 points; the two representations connected by dashed lines represent a weak complementary relationship between the two, with a corresponding complementarity score of 1 point.

[0096] Figure 3 This is a schematic diagram of a future self-conflict network in one embodiment of the system of the present invention, wherein the two representations connected by the dashed line represent that there is a weak conflict relationship between them, and the corresponding conflict score is 1 point. Detailed Implementation

[0097] 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.

[0098] To facilitate understanding of the technical solution of this invention, the invention will be further described in detail below with reference to specific application scenarios. It should be noted that the following scenarios are only used to illustrate the implementation process of this invention and do not constitute a limitation on the scope of protection of this invention.

[0099] This embodiment provides a multi-agent collaborative vocational assessment system. Please refer to [link / reference]. Figure 1 It includes an information collection module, a dedicated agent construction module, a relevance scoring module, a network construction module, a network analysis module, a job recommendation module, and a report generation module.

[0100] In this embodiment, the target user is a software technology major with a college diploma who wants to obtain a career development report tailored to their personal characteristics and development needs.

[0101] The information collection module is used to acquire users' demographic information, personality trait vectors, future representation sets, and user request texts.

[0102] In this embodiment of the invention, the information collection module includes a demographic information collection unit, a personality feature vector collection unit, a future representation set collection unit, and a user request text data collection unit.

[0103] The demographic information collection unit is used to acquire the demographic information of the target user. :

[0104] The system receives demographic information input by the target user through an interactive interface. This demographic information includes at least: user identity information, user gender, user education level, and user major. The user identity information can be an email address, phone number, or an automatically generated device ID, used to identify the user in the career development report. Preset options for user gender include male, female, and others. Preset options for user education level include high school and below, associate degree, bachelor's degree, master's degree, and doctoral degree. The user major is in free text format, recording the user's field of education.

[0105] The personality feature vector acquisition unit is used to acquire the personality feature vector of the target user. :

[0106] In this embodiment of the invention, the Big Five personality assessment data is obtained by calling a pre-set standardized Big Five personality scale. The scale contains five independent and complete psychological dimensions, and the psychological semantic definitions of each dimension are as follows:

[0107] Neuroticism (N): Characterizes the emotional stability and anxiety tendency of the target user;

[0108] Extraversion (E): Characterizes the sociality and activity level of the target user;

[0109] Openness to Experience (O): Represents the degree to which the target user is open to new things;

[0110] Agreeableness (A): Characterizes the target user's tendency to cooperate and trust with others;

[0111] Conscientiousness (C): Characterizes the target user's self-control and sense of responsibility.

[0112] The standardized Big Five personality traits scale has eight items under each dimension, and each item is scored using a five-point Likert scale. Specifically, each item has five ordered response options, corresponding to scores from 1 to 5: 1 represents "strongly disagree", 2 represents "disagree", 3 represents "neutral / uncertain", 4 represents "agree", and 5 represents "strongly agree".

[0113] After the target user completes all the questions in a certain dimension, the arithmetic mean of the scores for the 40 questions in that dimension is taken as the final score for that dimension. The final scores of the five dimensions constitute a personality trait vector. , ,in, For open-ended scoring, To score for due diligence, For extraversion scores, To score for agreeableness, Neuroticism score. Each score ranges from 1 to 5.

[0114] The future representation set acquisition unit is used to acquire the future representation set. :

[0115] We used the Future Representation Scale from Strauss et al. (2025), which includes 206 future-oriented cognitive representation items, categorized into the following 10 categories:

[0116] 1. Visual Appearance: The outward appearance of one's future self. Examples: "My future self is professionally dressed," "My future self is physically healthy."

[0117] 2. Social Categories: The group, circle, or identity category to which the future self belongs. Examples: "Belongs to a group of senior technical experts," "I identify as a member of a certain industry association," "I am a practitioner of environmental protection concepts."

[0118] 3. Behavior with Others: How the future self interacts with others in social, work collaboration, and other scenarios. Examples: "The future me will proactively coordinate cross-departmental collaborations" and "I will frequently communicate and learn from industry seniors."

[0119] 4. Relationships with Others: Stable connections (such as professional and personal relationships) that you will build with others in the future. Examples: "Become a trusted mentor to team members" and "Maintain long-term partnerships with key clients".

[0120] 5. Evaluation by Others: How others perceive, approve of, or provide feedback about you in the future, based on your expectations. Examples: "Being evaluated as a 'reliable partner'" or "Gaining professional recognition within the industry."

[0121] 6. Behavior Alone: ​​Activities, habits, or behavioral patterns of the future self when alone. Examples: "The future me will spend 1 hour every day studying new technologies" and "Regularly spend time alone to review work results."

[0122] 7. Affective Experiences: The emotional state that the future self will feel in their career or life. Examples: "Feeling a sense of accomplishment at work", "Maintaining calm and composure when facing challenges".

[0123] 8. Traits and Characteristics: Core attributes such as personality, abilities, and qualities that you will possess in the future. Examples: "I will have keen insight in the future," "I will become a patient and responsible leader."

[0124] 9. Experiencing (Imagined) Events: Specific scenarios or events that your future self might participate in or encounter. Examples: "Leading the implementation of an important project," "Attending and speaking at a top industry summit."

[0125] 10. Experiencing (Imagined) Surroundings: The physical space, work environment, or living environment in which your future self will be. Examples: "Working in an open-plan, innovative office," "Frequently traveling to different cities for business exchanges."

[0126] To achieve an accurate match between future self-representation and career position, this invention customizes 206 future-oriented cognitive representation items into 10 preset category attributes: skill acquisition, job promotion, education improvement, domain expansion, resource accumulation, network expansion, influence building, work-life balance, innovation ability, and stability pursuit.

[0127] Users select 4 to 20 future representations from the future representation scale, which are then standardized and packaged into a future representation set. , ,in, The total number of self-representations selected by the user.

[0128] The user request text data acquisition unit is used to acquire user request text data. The structure is {"Target Domain 1": response text, "Target Domain 2": response text, ..., "Target Domain 1": response text, ..., "Target Domain 2": response text, ..., "Target Domain 3": response text, ..., "Target Domain 4": response text, ..., "Target Domain 5": response text, ..., "Target Domain 6": response text, ..., "Target Domain 7": response text, ..., "Target Domain 8": response text, ..., "Target Domain 9": response text, ..., "Target Domain : Answer text}.

[0129] In this embodiment, open-ended questions from three target domains are presented to the user to collect unstructured semantic data from the user in different domains. After standardization processing, the user's request text data is obtained. It contains three key-value pairs with the structure {"Career Development": text, "Knowledge Learning": text, "Life Vision": text}.

[0130] The specific configuration for this issue is as follows:

[0131] Open-ended question 1 belongs to the career development field, and the question is "What issues are you most concerned about in your career development?"; open-ended question 1 is used to extract users' deep-seated needs in career development.

[0132] Open-ended question 2 belongs to the knowledge learning field, and the question content is "When it comes to your own knowledge learning, what issues are you most concerned about?"; Open-ended question 2 is used to extract users' deeper needs in knowledge learning.

[0133] Open-ended question 3 belongs to the life vision domain, and the question content is "What issues are you most concerned about when facing your life vision?"; open-ended question 3 is used to extract users' deep-seated needs regarding their life vision.

[0134] The standardization process described in the future representation set acquisition unit and the user request text data acquisition unit includes the following steps:

[0135] (1) Text cleaning: Remove special characters (such as HTML tags, emojis, control characters) from the text, convert full-width characters to half-width characters, and trim leading and trailing whitespace characters;

[0136] (2) Encoding standardization: uniformly convert to UTF-8 encoding format;

[0137] (3) Content validation (for future representations only): Check whether each representation text is a valid item in the 206 standard representation items of the Strauss scale, and filter out invalid or tampered inputs;

[0138] (4) Length control: The text of the answer to the open question is truncated, with a maximum length limit of L=500 characters. Any part exceeding this limit is discarded.

[0139] The Dedicated Agent Construction Module is used to build a user-specific future self-assessment agent based on the user's demographic information and personality trait vectors. The Dedicated Agent Construction Module includes a system prompt word construction unit, an Agent inference context configuration unit, a session instance creation unit, a validity verification unit, and a session instance generation unit.

[0140] System prompt word building units are used to construct structured system prompt words:

[0141] Utilizing demographic information from user structured personality profile data and personality trait vector Construct structured system prompts. The template format for structured system prompts is: "You are a career development assessment expert and now need to conduct an assessment from the perspective of the following users; Background information: Demographic information vector of the target users." Personality traits: the personality characteristic vector of the target user. Based on the user's background information and personality traits, please assess the correlation between the following two future representations.

[0142] The Agent inference context configuration unit is used to configure the Agent inference context:

[0143] By using the system parameter injection mechanism of the large language model API, the system prompt words are configured as the inference context of the user's exclusive future self-rating agent.

[0144] For example, a message list containing role="system" messages and role="user" messages is constructed, where the content field of the system message is the text of the system prompt word mentioned above, and the content field of the user message is the text of the future representation pair to be evaluated. By calling the dialogue completion interface of the large language model (such as the OpenAI Chat Completions API or a compatible interface), the model will use the user personality data in the system message as a contextual constraint during the inference process, and generate a rating result that conforms to the user's perspective.

[0145] The session instance creation unit is used to create a user-specific future self-scoring Agent session instance:

[0146] The system creates an independent dialogue session for each user, and the system prompts for that user are persistently stored in the system parameter of the session. All subsequent rating requests for that user's future representations are made by calling the Large Language Model API within this session context, ensuring that the rating behavior is continuously constrained by the user's personality profile.

[0147] The validity verification unit is used to verify the validity of the user-specific future self-scoring Agent configuration.

[0148] The system uses a pre-set set of test future representation pairs (containing 5 sets of future representation pairs with clear expected scoring relationships, such as becoming a team's technical leader and learning data science in the system are highly complementary, while frequent business trips to expand business and pursuing work-life balance are conflicting) to verify the effectiveness of the configured user-specific future self-scoring Agent.

[0149] The specific operation is as follows: input the test future representation pairs into the Agent sequentially, record the returned score values, and compare them with the preset expected score range. The user's personalized future self-scoring Agent configuration is considered verified when both of the following conditions are met.

[0150] Condition 1: The same test characterization pair in In each repeated call, the variance of the score value is less than the threshold. ;in, , .

[0151] Condition 2: The scoring direction (positive / negative / zero) of each test representation pair is consistent with the expectation.

[0152] If the first Agent configuration verification fails, return to the Agent inference context configuration unit and adjust the temperature parameter of the large language model API (from the default value of 0.7 to 0.3 to reduce randomness); if the second Agent configuration verification fails, return to the system prompt word construction unit, check the prompt word template format, reconfigure and verify again; repeat the above retry logic until verification passes, and then enter the session instance generation unit.

[0153] The session instance generation unit is used to generate user-specific representations of future self-scoring agent sessions.

[0154] After the Agent configuration is verified, an independent dialogue session instance is created for the current target user, and the session identifier and the user's system prompt word configuration are persistently stored in the session management module.

[0155] The user-specific representation of future self-rating Agent refers to reusing the same basic large language model and maintaining independent system parameter configurations and session contexts for each user to achieve differentiated rating behaviors when different users call the same model, rather than training or deploying model instances separately for each user.

[0156] The correlation scoring module is used to use the user's exclusive future self-scoring agent to score the complementarity of any two representations in the future representation set, and obtain the correlation score results of the pairwise representations.

[0157] Specifically, from the future representation set Any two different future representations can be selected to form a future representation pair, and all future representation pairs constitute the set of future representation pairs. .

[0158] For example, user's future representation data It includes four future representations, ,in, The content is "Become the team's technical lead"; The content is "Systematic Learning of Data Science"; The content is "achieving work-life balance"; The content is "Building personal brand influence".

[0159] From the four future representations, randomly select two different future representations to form an unordered pair, generating a set of representation pairs. .

[0160] The set of future representation pairs will be arranged sequentially. Each future representation pair and structured user cue words are input into the user-specific agent, and the user-specific representation future self-scoring agent outputs a relevance score for each future representation pair. , .

[0161] The relevance scoring uses a five-level scale to quantify the relevance between two future representations in a future representation pair. The five levels are: +2, +1, 0, -1, and -2. A positive relevance score indicates a complementary relationship between the representation pairs, while a negative score indicates a conflicting relationship. The absolute value of the score represents the strength of the effect; the larger the absolute value, the stronger the complementarity or conflict. A score of 0 indicates no significant correlation between the representation pairs. Specifically, +2 indicates that one future representation strongly promotes the realization of the other; +1 indicates that one future representation positively promotes the realization of the other; 0 indicates that the two future representations do not affect each other; -1 indicates that one future representation negatively hinders the realization of the other; and -2 indicates that one future representation strongly hinders the realization of the other.

[0162] Establish a mapping relationship between future representations and scores, wherein the mapping relationship satisfies a two-way symmetry constraint, that is, the influence score of a future representation B on a future representation B is equal to the influence score of a future representation B on a future representation A.

[0163] This embodiment represents the mapping relationship between future representations and scores by constructing a relevance scoring matrix. Specifically, the relevance scoring matrix is ​​constructed. Relevance score matrix Both rows and columns correspond to the future representation set. Each future representation element in the matrix; correlation score matrix Diagonal elements are defined as empty or zero, indicating that future representations do not interact with themselves; correlation score matrix elements Representation of the future With future representation The correlation scores between them satisfy symmetry. ;in, and Index for future characterization , , .

[0164] For example, please refer to Table 1, which is a set of future representation pairs. The corresponding relevance score matrix, where "--" indicates a null value.

[0165] Table 1

[0166]

[0167] The network building module is used to construct complementary networks and conflict networks based on the correlation scores of any two representations.

[0168] In this module, the method for constructing complementary networks is as follows:

[0169] Using future representations in the future representation set as nodes of the complementarity network, if the correlation score between two future representations is greater than or equal to a first preset threshold, an undirected edge is established between the two future representations, with the edge weight being the correlation score result; no edge is established between representations whose correlation score is less than the first preset threshold.

[0170] In this embodiment, A future representation as a complementary network There are 1 node; the first preset threshold is set to +1, and an undirected edge is established between the representation pairs corresponding to the correlation score results of +1 or +2, with the edge weight being the correlation score result; no edge is established between the representation pairs corresponding to the correlation score results of 0, -1 or -2.

[0171] For example, refer to Figure 2 , Figure 2 This is a schematic diagram of a complementary network structure.

[0172] In this module, the method for constructing conflict networks is as follows:

[0173] Using future representations in the future representation set as nodes of the complementarity network, if the correlation score between two future representations is less than or equal to a second preset threshold, an undirected edge is established between the two future representations, with the edge weight being the absolute value of the correlation score; no edge is established between representations whose correlation score is greater than the second preset threshold.

[0174] In this embodiment, A future representation as a conflict network There are 1 node; the second preset threshold is set to -1, and an undirected edge is established between the representation pairs corresponding to the correlation score results of -1 or -2, with the edge weight being the absolute value of the correlation score result; no edge is established between the representation pairs corresponding to the correlation score results of 0, +1 or +2.

[0175] For example, refer to Figure 3 , Figure 3 This is a schematic diagram of a conflict-prone network structure.

[0176] The network analysis module is used to label hub nodes and delineate network community structures for complementary and conflicting networks, respectively.

[0177] In this module, the Personality-Weighted Centrality Index (PWCI) of all nodes in complementary or conflict networks is calculated, and nodes with a PWCI value greater than a third threshold are selected. The nodes form a candidate Hub node set. Determine if the set of candidate Hub nodes is not empty:

[0178] If the candidate Hub node set If the set of candidate Hub nodes is not empty, then the nodes in the set of candidate Hub nodes will be determined as Hub nodes.

[0179] If the candidate Hub node set If the value is empty, all nodes are sorted in descending order based on the PWCI value, and the node with the highest sorting position (Num) is selected as the Hub node.

[0180] In a complementary network, a Hub node represents a core anchor point for career development weighted by a user's personality traits. For the same complementary network, different users may identify different Hub nodes.

[0181] The formula for calculating the Personality Weighted Centrality Index (PWCI) is as follows:

[0182] (1)

[0183] in, For nodes in complementary or conflicting networks, , For a set of nodes; For nodes PWCI value; For nodes The set of neighboring nodes, representing the relationship with the node. There are all other future representations that are connected. ; For the edge The weights; For nodes The corresponding preset representation type; For user personality feature vectors; The personality-type fit function is used to calculate the user's personality feature vector. With preset representation type The degree of compatibility.

[0184] Personality-Type Compatibility Function The expression is:

[0185] (2)

[0186] in, It is a collection of the Big Five personality traits. , As an open dimension, For the due diligence dimension, For the extroversion dimension, For the human-centered dimension, Neuroticism dimension; a set of the Big Five personality dimensions The corresponding Big Five personality traits score vectors serve as the user's personality trait vectors. , ,in, For open dimension Score, For the dimension of responsibility Score, For the extroversion dimension Score, For the human-friendly dimension Score, Neuroticism dimension Score; Represents the set of Big Five personality dimensions Index; For dimension Score Normalization to The standardized score after that; Preset dimensions The global weights represent the sensitivity of different representation types to different dimensions. The values ​​are based on the presuppositions of occupational psychology experimental research and satisfy... , express The absolute value; For dimension The standardized score baseline threshold, by default In this embodiment, the corresponding original score is 3; For dimension For preset representation types Adjustment direction, , When, it represents dimension. With representation type Positive correlation; When, it represents dimension. With preset representation type Negative correlation.

[0187] In this embodiment, the third threshold Set as ,in The mean of the PWCI values ​​for all nodes. is the standard deviation of the PWCI values ​​for all nodes; Num is 3.

[0188] Based on the hub node identification method of complementary networks, those skilled in the art can derive the hub node identification method of conflict networks without creative effort, so this paper will not elaborate further.

[0189] This module employs the Personality-Aware Louvain (PAL) algorithm to partition the network community structure of complementary and conflicting networks. Specifically, it includes a first-stage community partitioning and a second-stage network aggregation. The details are as follows:

[0190] Phase One Community Division:

[0191] (1) Initial partitioning

[0192] Each node in the current network to be partitioned is initialized as an independent community, such that the number of communities is the same as the number of nodes in the initial state, and each community contains only a single node. On the first execution, the current network to be partitioned is either a complementary network or a conflicting network; on subsequent executions, the current network to be partitioned is the aggregated network obtained in the second stage.

[0193] (2) Iterative optimization

[0194] Repeat the following sub-steps until the first stage terminates:

[0195] (2.1) Node traversal

[0196] In the current iteration, each node in the network is traversed in a random order. Each node is traversed and processed only once in the same iteration.

[0197] (2.2) Candidate Community Determination

[0198] For the current node Determine the node A set of candidate communities; the set of candidate communities includes: nodes The current community and its relationship with nodes The community to which neighboring nodes with connections belong.

[0199] (2.3) Calculation of modularity gain

[0200] For each candidate community in the candidate community set, calculate the node The increment in modularity caused by migrating to a candidate community. Where the candidate community is a node... When the current community is occupied, the corresponding module degree increment value is zero.

[0201] (2.4) Optimal Community Selection

[0202] Compare the module degree increment values ​​of each candidate community, and select the candidate community with the largest module degree increment value as the target community. .

[0203] (2.5) Node migration determination

[0204] Determine if the maximum modularity increment is greater than zero: if so, then add the node. Migrate from the current community to the target community If not, then the node Remain in the original community and do not relocate.

[0205] (2.6) Termination of the first stage

[0206] After completing the traversal of all nodes in the current round, check whether any changes in node community affiliation occurred during this round of traversal:

[0207] If the community affiliation of at least one node changes, then based on the updated community division, return to the node traversal sub-step (2.1) and continue execution;

[0208] If no node's community affiliation changes during this round, the process is considered converged, the first phase ends, and preliminary community division results are obtained.

[0209] In this module, the formula for calculating the module degree gain is:

[0210] (3)

[0211] in, The personality perception modularity gain represents the change in network modularity after node migration. This represents the total edge weight of the network, and its value is half of the sum of the edge weights of all edges in the currently partitioned network. For nodes and nodes Edge weights between them; For nodes The weighting degree, For nodes The weighted degree, which is defined as the sum of the weights of all edges associated with the current node; For community indicator functions, when node and nodes The value is 1 if the two communities are in the same area, and 0 otherwise. For nodes Current community For nodes Current community affiliation; For nodes With nodes The personality similarity coefficient between individuals, which integrates personality compatibility and type association, is expressed as follows:

[0212] (4)

[0213] in, and For personality-type fit function, User personality feature vector With nodes Corresponding preset representation type The degree of compatibility, User personality feature vector With nodes Corresponding preset representation type The degree of compatibility; Preset representation type With preset representation type The degree of correlation.

[0214] Personality-Type Compatibility Function and The expression is:

[0215] (5)

[0216] (6)

[0217] in, Preset representation type Big Five personality dimensions Weight on, Preset representation type Big Five personality dimensions Weights on, weights and weight The weights are determined according to the preset weight table; please refer to Table 2.

[0218] Table 2 Preset Weight Table

[0219]

[0220] The preset correlation degree described in this embodiment is obtained by taking the value of the following correlation matrix, please refer to Table 3.

[0221] Table 3

[0222]

[0223] After the first phase, a preliminary community partitioning scheme was obtained, in which each node in the network was assigned to a different community. Nodes within the same community were highly correlated in terms of personality perception similarity and edge weights.

[0224] Second-stage network aggregation:

[0225] After completing the first phase of community partitioning optimization, network aggregation is performed, which includes:

[0226] (1) Constructing supernodes

[0227] Each community identified in the first phase is abstracted as a supernode, and all supernodes constitute the node set of the aggregated network.

[0228] (2) Define the edge weights between supernodes

[0229] For any two distinct supernodes and Define the edge weights between the two. for:

[0230] (7)

[0231] (3) Construct the aggregation network and trigger the next iteration.

[0232] Based on the set of supernodes and the edge weights between them, an undirected weighted aggregation network is constructed. This aggregation network is then used as the new input network, and the process returns to the first stage of community partitioning to continue.

[0233] Repeat the first phase of community partitioning and the second phase of network aggregation until one of the following conditions is met:

[0234] Condition 1: The results of the first phase of community delineation are exactly the same in two consecutive phases;

[0235] Condition 2: The current community structure division rounds have reached the preset maximum round threshold.

[0236] When terminating the community structure partitioning, the community structure in the currently obtained aggregated network is determined as the final community partitioning result of either a complementary network or a conflict network.

[0237] It should be noted that the first-stage community division and the second-stage network aggregation are both divisions of node affiliation relationships in complementary or conflicting networks, and do not constitute modifications to the node set, edge set, or edge weight of the complementary or conflicting network; the graph structure of the complementary or conflicting network remains unchanged throughout the entire iteration process.

[0238] The job recommendation module is used to invoke multiple intelligent agents with different functions, based on user personality feature vectors. Future representation set Complementary networks and conflict networks are used to automatically generate job recommendation results through multi-agent collaborative reasoning.

[0239] In this module, the multi-agent system includes four pre-built specialized agents: the first agent, the second agent, the third agent, and the fourth agent.

[0240] In this embodiment, the first intelligent agent is a personality analysis intelligent agent, the second intelligent agent is a representation analysis intelligent agent, and the third intelligent agent is a job matching intelligent agent and a report integration intelligent agent.

[0241] The four agents are functionally decoupled and constructed based on the same pre-trained large language model instance. Each agent possesses different domain expertise and task boundaries by configuring differentiated system prompts. For details, please refer to Table 4.

[0242] Table 4

[0243]

[0244] The execution process for generating job recommendation results collaboratively by various intelligent agents includes the following steps:

[0245] Step S601: Invoke the report integration agent and issue instructions containing task type, input data and return data format requirements to the personality analysis agent, representation analysis agent and job matching agent.

[0246] The instructions issued to the personality analysis agent specify the task type as: interpretation of the Big Five personality traits across its five dimensions and generation of career matching suggestions; the input data is the user's personality feature vector. The returned data must be structured. A template example is: {Interpretation of scores for each dimension: {Openness: text, Conscientiousness: text, Extraversion: text, Agreeableness: text, Neuroticism: text}, Dimensional interaction analysis: text, Personality profile description: text, Career matching suggestions: text}.

[0247] The instructions issued to the representation analysis agent specify the task type as: future representation network topology analysis and group identification; the input data is: a set of future representations. Complementary networks and conflict networks; the returned data must be structured, and a template example is: {Hub node identification: text, representation group analysis: text, complementarity analysis: text, conflict analysis: text, representation priority sorting: [representation list]}.

[0248] The instructions issued to the job matching agent specify the task type as: calculating job matching scores based on personality traits and representation groups; the input data consists of data output by the personality analysis agent and data output by the representation analysis agent; the return format is required to be structured data, with a template example as: {Recommended Job List:[{Job Name:Text, Matching Score:Value, Matching Reason:Text}], Highlighted Trait Combinations:Text}.

[0249] Step S602: The personality analysis agent, the representation analysis agent, and the job matching agent execute in parallel based on instructions, and each generates analysis data.

[0250] The processing steps of a personality analysis agent include: receiving personality feature vectors. It generates score interpretations for each dimension, dimension interaction analysis (i.e., the impact analysis of different combinations of the Big Five personality dimensions on career choices, such as the impact analysis of the combination of openness and conscientiousness on career choices), personality profile descriptions, and career matching suggestions, and encapsulates the results into structured data output.

[0251] The process of characterization analysis of intelligent agents includes:

[0252] Receive future representation set Complementary networks and conflict networks;

[0253] Identify all Hub nodes in complementary and conflicting networks as the corresponding text content for "Hub Node Identification"; treat each community in the complementary and conflicting networks as a career development theme, and all communities constitute a career development theme group. Based on the career development theme group, obtain the corresponding text content for "Representation Group Analysis"; analyze the facilitating relationships between the corresponding representations of each node according to the complementary network analysis as the corresponding text content for "Complementarity Analysis"; analyze the hindering relationships between the corresponding representations of each node according to the conflicting network analysis as the corresponding text content for "Conflict Analysis"; use the Personality Weighted Centrality Index (PWCI) to analyze the future representation set. All future representations are sorted in descending order. When multiple representations have the same PWCI value, they are sorted in descending order according to the size of the community to which the representation belongs. The community size is defined as the total number of representations contained in the corresponding community. The sorting result is used as the representation list content corresponding to "representation priority sorting". The above processing results are encapsulated into structured data output.

[0254] The processing steps of the job matching agent include:

[0255] (1) Receive the personality analysis results generated by the personality analysis agent and the representation analysis results generated by the representation analysis agent.

[0256] (2) Based on the two personality dimensions with the highest scores in the user personality analysis results, select job positions in the preset job database whose personality dimension tags cover the two personality dimensions with the highest scores to form a candidate job set.

[0257] (3) For each position in the candidate position set, calculate the vector similarity index, network coverage index and path consistency index respectively.

[0258] The vector similarity index is calculated based on the personality analysis results, using the following formula:

[0259] (8)

[0260] in, For the position The required Big Five personality score threshold vector; The Big Five personality dimensions The job-specific weights represent the differences in sensitivity of different jobs to each dimension. Their values ​​can be determined by referring to the correlation between different occupations and each dimension of the Big Five personality traits in Anni, K., Vainik, U., & Mõttus, R., "Personality profiles of 263 occupations", Journal of Applied Psychology, 2025, 110(4): 481-497. For the current user in the dimension The standardized score is denoted as . ; For the position Requirements in dimension The scoring threshold is based on a preset value in the job database. .

[0261] The network coverage index is calculated based on the characterization analysis results, using the following formula:

[0262] (9)

[0263] in, For the position Network coverage metrics for complementary networks; For the position The required set of core self-representations is pre-defined based on the job database; For complementary networks The set of Hub nodes; For set With sets The cardinality of the intersection, i.e., the set of core representations of the job requirements. The set of user hub nodes The number of representations; For set The cardinality (number of elements), i.e., the job position The required total number of core representations; For set With sets Elements in the intersection; For Hub node set The elements in.

[0264] The path consistency index is calculated based on the characterization analysis results, using the following formula:

[0265] (10)

[0266] in, For the position With conflict networks Path consistency; For conflict networks In the process, from any Hub node in the Hub node set to the job position. The set of edges covered by all shortest paths in the core representation; For conflict networks In the process, from any Hub node in the Hub node set to the job position. The set of edges covered by all feasible paths representing the core representation; For the edge The weights, where , For conflict networks Nodes and edges in Belongs to set ; For conflict networks Middle The weight, and For conflict networks Nodes and edges in Belongs to set The higher the path consistency, the better the job position. The lower the intensity of conflict with users.

[0267] (4) By linearly weighting and fusing the vector similarity, network coverage and path consistency indicators, a matching score between each job and the user is generated.

[0268] The formula for calculating the matching score between the job and the user is as follows:

[0269] (11)

[0270] in, For candidate positions With the current user Match score; , and The three-dimensional weight coefficients satisfy the following conditions: Preferred value , , ; This is a vector similarity index function; This is a function representing network coverage metrics. This is a path consistency index function.

[0271] (5) Sort all positions in the candidate position set in descending order of position matching score, and select the top 5 positions to form a recommended position list.

[0272] (6) For each job in the recommended job list, generate an interpretable textual reason for matching based on similarity index, network coverage index and path consistency index.

[0273] (7) Encapsulate the recommended job list, the matching score of each job and the matching reason into structured data in a predefined format and output it to the report integration agent.

[0274] The report generation module is used to generate personalized career development reports based on job recommendation results and user request texts, following a preset template.

[0275] In this module, the report integration agent, based on a preset report template, integrates data generated by the personality analysis agent, the representation analysis agent, and the job matching agent to generate a career development report.

[0276] Specifically, the following steps are included:

[0277] Step S701: After receiving the structured data returned by the personality analysis agent, the representation analysis agent, and the job matching agent, the report integration agent performs a three-level verification mechanism:

[0278] Field integrity validation: Checks whether the returned data contains the pre-defined required fields. Among them,

[0279] The output data of the personality analysis agent should include four fields: dimensional score interpretation, dimensional interaction analysis, personality profile description, and career matching suggestions; the output data of the representation analysis agent should include five fields: Hub representation identification, representation group analysis, complementarity analysis, conflict analysis, and representation priority ranking; the output data of the job matching agent should include four fields: user's prominent trait combination, recommended job list, matching score, and matching reason.

[0280] Data type validation: Checks whether the data type of each field conforms to the preset specifications. For example, the matching score must be a numeric value within the range of 0-100, and the recommended job list must be an array.

[0281] Reference consistency check: Check whether the representation identifiers, dimension names, etc. referenced by each agent are consistent with the original data source generated in the previous steps.

[0282] Step S702: If any of the above verification items fails, the integrated agent shall handle the report according to the following hierarchical strategy:

[0283] Initial verification failed: The report integrates the agent and sends a retry instruction to the corresponding agent, carrying structured error description information in the instruction, such as missing dimension interaction analysis field or matching score value exceeding the valid range, requiring the agent to complete the correction and return the structured result within no more than three inference cycles.

[0284] Failed duplicate validation: If the same agent fails validation after three consecutive retries in the same data request context, the integrated agent is notified to activate the preset fault tolerance mechanism.

[0285] For missing or abnormal fields, use context-compatible default values ​​to fill them in (for example, when personality analysis data is missing, call the median interpretation template for each dimension of the Big Five personality traits to complete it; when job matching data is missing, generate an empty list and return a list marked "No recommended jobs available").

[0286] At the same time, abnormal events and their context information are recorded in the error log for subsequent manual auditing and model iteration optimization.

[0287] The above fault-tolerance strategy ensures that the report generation process is not interrupted due to an abnormal output from a single agent.

[0288] Step S703: After successful verification, the reporting and integration agent performs the data integration operation:

[0289] The structured outputs from three specialized intelligent agents are integrated into a unified data object; semantic reference relationships are established between fields within the data object; for example, personality traits in job matching results are linked with corresponding dimension descriptions in personality analysis results.

[0290] Step S704: Generate a personalized career development report

[0291] The career development report contains five chapters:

[0292] Chapter 1: User Overview

[0293] It presents the user's basic information, a radar chart of quantitative scores for the Big Five personality dimensions, and a personalized cartoon image. The personalized cartoon image is automatically generated by constructing image-generating prompts based on the user's personality profile description, Hub representation, representation group analysis results, and the Top-1 job recommended by the job matching agent, and then calling a generative artificial intelligence image model to render and generate it.

[0294] Chapter Two: In-depth Analysis of Personality Traits

[0295] Based on the structured data output by the personality analysis agent, the system interprets the user's performance in each dimension of the Big Five personality traits; and provides career development suggestions and interpersonal interaction style analysis that are highly relevant to each dimension.

[0296] Chapter 3: Future Representation Network Analysis

[0297] Based on the structured data output by the representation analysis agent, the topology of complementary and conflicting networks is displayed in the form of a visual graph, and the identified Hub representation nodes (core occupational anchors) and representation groups are marked in the complementary network graph.

[0298] Chapter Four: Career Path Recommendations

[0299] Based on the structured data output by the job matching agent, recommended career directions and matching reasons are displayed in list form.

[0300] Chapter 5: Personalized Learning Recommendations and Development Suggestions

[0301] Based on users' core career anchors and complementary network community structures, the report identifies users' core learning directions and recommends online courses and books for each direction. Furthermore, based on users' answers to three open-ended questions, the report integrates an intelligent agent to generate personalized, in-depth responses, providing customized suggestions addressing users' deeper needs in career development, knowledge acquisition, and life aspirations.

[0302] Chapter Six: Personalized Conclusion

[0303] The report integrates the user's personality traits, core career anchors, and open-ended answers to questions to generate a personalized conclusion with a warm and encouraging tone. Written in a friendly tone, the conclusion affirms the user's focus and investment in their career development, conveying positive emotional values.

[0304] In summary, the present invention has the following advantages over the prior art:

[0305] First, by using user background information and Big Five personality traits as system prompts for the Agent, a personalized scoring mechanism of "one person, one Agent" is constructed, capable of replacing the user's own scoring. This Agent does not acquire scoring ability through training, but rather by injecting system prompts into the user profile, enabling the large language model to make complementary judgments from the user's perspective. Compared to traditional rule-matching methods that can only process preset rules, the Agent of this invention can understand the semantic content of free text and provide differentiated scores based on the user's personalized psychological traits. For example, given the same two characteristics—"becoming the team's technical lead" and "systematically learning data science"—the Agent for a highly conscientious user will provide a very helpful score (learning and promotion are synergistic under self-discipline), while the Agent for a low-conscientious user will provide a helpful score (insufficient learning execution may slow down the promotion process).

[0306] Second, by constructing future self-complementarity networks and future self-conflict networks respectively, referencing the network method of Strauss et al. (2025), we can systematically reveal the mutually reinforcing and mutually hindering relationships between users' career goals. Existing technologies can only analyze each career goal in isolation or make simple positive and negative correlation classifications, while this invention calculates complementarity and conflict scores based on the weighted density of network science, identifies users' core career anchors through Hub node analysis, and discovers career development theme groups through community structure detection. From a global topological perspective, it grasps users' career development demands and achieves quantitative modeling of the relationships between individuals' future self-representations.

[0307] Third, through a multi-agent collaborative system, intelligent decomposition and parallel processing of the report generation task were achieved. The three specialized agents—personality analysis, representation analysis, and job matching—each perform their respective duties and operate simultaneously, significantly shortening the report generation time compared to the traditional serial processing mode. The report integration agent coordinates the results of each agent to ensure the logical consistency and structural integrity of the report.

[0308] Fourth, by integrating psychological assessment, agent system prompt word configuration, neural network analysis, multi-agent collaboration, and generative artificial intelligence technologies, an end-to-end career development report generation pipeline has been constructed, realizing full automation from user data collection to report output. It can provide high-quality and personalized career development planning services to a large number of users without human intervention.

[0309] In summary, this invention, by combining a multi-agent collaborative architecture with a future self-assessment mechanism, effectively overcomes the technical bottlenecks of existing career planning systems in semantic understanding, relationship modeling, and data fusion, thereby improving the automation and personalization of career development path planning.

[0310] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0311] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-agent collaborative vocational assessment system, characterized in that, include: The information collection module is used to acquire users' demographic information, personality trait vectors, future representation sets, and user request texts. The Dedicated Agent Construction Module is used to construct structured system prompts for a user's dedicated future self-rating agent based on demographic information and personality feature vectors, and to configure the system prompts as the inference context for the user's dedicated future self-rating agent. The correlation scoring module is used to use the user's exclusive future self-scoring agent to score the correlation between any two representations in the future representation set, and obtain the correlation score results of each pair of representations. The network construction module is used to construct complementary networks and conflict networks based on the relevance scoring results, respectively. The network analysis module is used to identify hub nodes and classify community structures for complementary and conflicting networks, respectively. The job recommendation module is used to invoke multiple intelligent agents with different functions to perform multi-agent collaborative reasoning based on personality feature vectors, future representation sets, complementary networks, and conflict networks to generate job recommendation results. The report generation module is used to generate personalized career development reports based on job recommendation results and user request texts, following a preset template.

2. The system according to claim 1, characterized in that, The method for constructing the complementary network is as follows: Using future representations in the future representation set as nodes of the complementary network, if the correlation score between two future representations is greater than or equal to a first preset threshold, an undirected edge is established between the two future representations, with the edge weight being the absolute value of the correlation score. No edges are established between representations whose correlation scores are less than the first preset threshold.

3. The system according to claim 1, characterized in that, The method for constructing the conflict-prone network is as follows: Using future representations in the future representation set as nodes of the complementarity network, if the correlation score between two future representations is less than or equal to a second preset threshold, an undirected edge is established between the two future representations, with the edge weight being the absolute value of the correlation score; no edge is established between representations whose correlation score is greater than the second preset threshold.

4. The system according to claim 2 or 3, characterized in that, The specific method for identifying the Hub node is as follows: Those selected were those with a Personality Weighted Centrality Index (PWCI) value greater than the third threshold. The nodes form a candidate Hub node set. Determine the set of candidate Hub nodes. Is it non-empty? If the candidate Hub node set If the set of candidate Hub nodes is not empty, then the nodes in the set of candidate Hub nodes will be determined as Hub nodes. If the candidate Hub node set If the value is empty, all nodes are sorted in descending order based on the PWCI value, and the node with the highest sorting position (Num) is selected as the Hub node.

5. The system according to claim 4, characterized in that, The expression for the Personality Weighted Centrality Index (PWCI) is as follows: ； in, As nodes in a complementary network, , A set of nodes for a complementary network; For nodes PWCI value; For nodes The set of neighboring nodes, representing the relationship with the node. There are all other future representations that are connected. ; For the edge The weights; For nodes The corresponding preset representation type; For user personality feature vectors; The personality-type fit function is expressed as follows: ； in, It is a collection of the Big Five personality traits. , As an open dimension, For the due diligence dimension, For the extroversion dimension, For the human-centered dimension, Neuroticism dimension; ,in, For open dimension Score, For the dimension of responsibility Score, For the extroversion dimension Score, For the human-friendly dimension Score, Neuroticism dimension Score; Represents the set of Big Five personality dimensions Index; For dimension The score was normalized to The standardized score after that; For dimension The global weight satisfies , express The absolute value; For dimension The standardized score baseline threshold; For dimension For preset representation types Adjustment direction, , When, it represents dimension. With representation type Positive correlation; When, it represents dimension. With preset representation type Negative correlation.

6. The system according to claim 5, characterized in that, The community structure partitioning of complementary and conflicting networks includes a first-stage community partitioning and a second-stage network aggregation, specifically as follows: Phase One Community Division: (1) Initial partitioning Each node in the current network to be partitioned is initialized as an independent community, such that the number of communities is the same as the number of nodes in the initial state, and each community contains only a single node; on the first execution, the current network to be partitioned is either a complementary network or a conflicting network; on subsequent executions, the current network to be partitioned is the aggregated network obtained in the second stage. (2) Iterative optimization Repeat the following sub-steps until the first stage terminates: (2.1) Node traversal In the current iteration, each node in the network is traversed in a random order; (2.2) Candidate Community Determination For the current node Determine the node A set of candidate communities; The candidate community set includes: nodes The current community and its relationship with nodes The communities to which neighboring nodes with connections belong; (2.3) Calculation of modularity gain For each candidate community in the candidate community set, calculate the node The increment in modularity caused by migrating to a candidate community; where the candidate community is a node. When the current community is involved, the corresponding module degree increment value is zero; The formula for calculating the modularity gain is: ； in, , where is the modularity gain, represents the change in network modularity after a node migration; This represents the total edge weight of the network, and its value is half of the sum of the edge weights of all edges in the currently partitioned network. For nodes and nodes Edge weights between them; For nodes The weighting degree, For nodes The weighted degree, which is defined as the sum of the weights of all edges associated with the current node; For community indicator functions, when node and nodes The value is 1 if the two communities are in the same area, and 0 otherwise. For nodes Current community For nodes Current community affiliation; For nodes With nodes The coefficient of personality similarity between them; (2.4) Optimal Community Selection Compare the module degree increment values ​​of each candidate community, and select the candidate community with the largest module degree increment value as the target community. ; (2.5) Node migration determination Determine if the maximum modularity increment is greater than zero: if so, then add the node. Migrate from the current community to the target community If not, then the node Remain in the original community; no relocation will occur. (2.6) Iteration Termination Determination After completing the traversal of all nodes in the current round, check whether any changes in the community affiliation of nodes occurred during this round of traversal: If the community affiliation of at least one node changes, then based on the updated community division, return to the node traversal sub-step (2.1) and continue execution; If no node's community affiliation changes during this round, the process is considered converged, the first phase ends, and preliminary community division results are obtained. Second-stage network aggregation: After completing the first phase of community partitioning optimization, network aggregation is performed, which includes: (1) Constructing supernodes Each community identified in the first phase is abstracted as a super node, and all super nodes constitute the node set of the aggregated network. (2) Define the edge weights between supernodes For any two distinct supernodes and Define the edge weights between the two. for: ； in, For nodes and nodes Edge weights between them; For nodes and nodes The coefficient of personality similarity between them; (3) Construct the aggregation network and trigger the next iteration. Based on the set of supernodes and the edge weights between them, an undirected weighted aggregation network is constructed; the aggregation network is then used as a new input network, and the process returns to the first stage of community partitioning to continue. Repeat the first phase of community partitioning and the second phase of network aggregation until one of the following conditions is met: Condition 1: The results of the first phase of community delineation are exactly the same in two consecutive phases; Condition 2: The current community structure division rounds have reached the preset maximum round threshold; When terminating the community structure partitioning, the community structure in the currently obtained aggregated network is determined as the final community partitioning result of either a complementary network or a conflict network.

7. The system according to claim 6, characterized in that, The expression for the personality similarity coefficient is: ； ； ； in, User personality feature vector With nodes Corresponding preset representation type The degree of compatibility, User personality feature vector With nodes Corresponding preset representation type The degree of compatibility; Preset representation type With preset representation type The degree of correlation; Preset representation type In dimensions Weight on, Preset representation type In dimensions Weighting on.

8. The system according to claim 7, characterized in that, The multiple intelligent agents with different functions are specifically: a first intelligent agent, a second intelligent agent, a third intelligent agent, and a fourth intelligent agent; The first intelligent agent issues instructions containing the task type, input data, and return data format requirements to the second, third, and fourth intelligent agents; The second, third, and fourth intelligent agents execute instructions in parallel, each generating analysis data: The second intelligent agent obtains personality analysis results based on personality feature vectors; The third agent obtains representation analysis results based on future representation sets, complementary networks, and conflict networks; The fourth agent generates job recommendation results based on personality analysis and representation analysis results; The job recommendation results include a list of recommended jobs, a matching score for each job, and the reasons for the matching.

9. The system according to claim 8, characterized in that, The fourth agent's score generates a job recommendation list and job matching degree using the following methods: Based on the two personality dimensions with the highest scores in the user personality analysis results, positions whose personality dimension tags cover the two highest-scoring personality dimensions are selected from a pre-set position database to form a candidate position set. For each position in the candidate position set, vector similarity index, network coverage index, and path consistency index are calculated. The vector similarity, network coverage, and path consistency indices are linearly weighted and fused to generate a position matching score between each position and the user. All positions in the candidate position set are sorted in descending order of matching score, and the top Num2 positions are selected to form a recommended position list.

10. The system according to claim 9, characterized in that, The expression for the vector similarity is: ； in, For the position The required Big Five personality score threshold vector; The Big Five personality dimensions The job-specific weights represent the differences in sensitivity of different jobs to each dimension; For the current user in the dimension The standardized score is denoted as . ; For the position Requirements in dimension The scoring threshold is based on a preset value in the job database. ; The expression for the network coverage metric is: ； in, For the position Complementary networks Network coverage metrics; For the position The required set of core self-representations is pre-defined based on the job database; For complementary networks The set of Hub nodes; For set With sets The cardinality of the intersection, i.e., the set of core representations of the job requirements. The set of user hub nodes The number of representations; For set The cardinality (number of elements), i.e., the job position The required total number of core representations; For set With sets Elements in the intersection; For Hub node set Elements in; The expression for the path consistency index is: ； in, For the position With conflict networks Path consistency; For conflict networks In the process, from any Hub node in the Hub node set to the job position. The set of edges covered by all shortest paths in the core representation; For conflict networks In the process, from any Hub node in the Hub node set to the job position. The set of edges covered by all feasible paths representing the core representation; For the edge The weights, where , For conflict networks Nodes and edges in Belongs to set ; For conflict networks Middle The weight, and For conflict networks Nodes and edges in Belongs to set .

Images