CN-122022105-A - Full-flow academic employment guidance system and method based on large model and multi-mode data

CN122022105ACN 122022105 ACN122022105 ACN 122022105ACN-122022105-A

Abstract

The invention relates to the technical field of application of artificial intelligence in education management, and discloses a full-flow academic employment guidance system and method based on a large model and multi-modal data, wherein the system collects the multi-modal data and reduces a vitamin feature set, and establishes real-time capability feature vector space coordinates through the large model; the judgment module calculates the deviation amount of the coordinate and the industry reference anchor point, and compares the deviation amount with a dynamic sensitivity threshold value to generate a trigger signal; the invention effectively suppresses behavioral noise interference through characteristic deviation triggering and reference dynamic correction, realizes cross-time domain alignment of a post competence boundary and an individual growth track, provides high-frequency perception and accurate prediction under low computational redundancy, and solves the phase deviation of talent culture and market variation.

Inventors

XIE QUANYING
LIN PINGFENG
YANG GUANGYUAN
ZHANG ZHEN
RAN BOWEN
Teng Qingwu

Assignees

西南民族大学

Dates

Publication Date: 20260512
Application Date: 20260414

Claims (10)

1. The full-flow academic employment guidance system based on the large model and the multi-mode data is characterized by comprising a data acquisition module, a feature mapping module, a feature deviation judging module and a path reconstruction module: The data acquisition module is used for acquiring multi-modal academic performance data aiming at a target individual, and carrying out semantic sampling and feature dimension reduction on the multi-modal academic performance data so as to generate an input feature set consisting of dominant business keywords and implicit capability features; The feature mapping module is used for inputting the dominant business keywords and the recessive capability features into the large model to obtain real-time capability feature vectors representing target individuals, and determining coordinate data of the real-time capability feature vectors in a preset feature space; The characteristic deviation judging module is used for calculating the characteristic deviation between the real-time capacity characteristic vector and a preset industry reference anchor point and generating a trigger signal according to the comparison result of the characteristic deviation and a sensitivity threshold, wherein the sensitivity threshold is a dynamic variable set according to the historical prediction accuracy; The path reconstruction module is used for responding to a trigger signal to execute the following path correction steps of S1, calling decision metadata comprising user expectations, the employment data of the present university, industry post increment rate, policy support weight, technical iteration rate and the like to construct a dynamic correction matrix representing industry development trend, S2, using the dynamic correction matrix as a linear predictor to conduct weighted offset calculation on a preset industry reference anchor point to obtain a dynamic prediction anchor point aligned with a target prediction time domain, and S3, regenerating a employment guidance path for a target individual according to a spatial distribution residual error between a real-time capability feature vector and the dynamic prediction anchor point.
2. The full-flow academic employment guidance system based on the large model and the multi-mode data according to claim 1 is characterized in that the characteristic deviation judging module is further used for carrying out trend verification on a trigger signal to improve the trigger reliability of path reconstruction, the trend verification comprises the steps of extracting characteristic deviation amounts according to a preset sampling period to establish a time axis evolution sequence, carrying out second-order differential calculation on the time axis evolution sequence to obtain a deviation evolution track of a target individual, calculating a transient slope value of the deviation evolution track, taking the product of the transient slope value and a preset semantic inertia weight as a corrected characteristic deviation amount, replacing the original characteristic deviation amount by the corrected characteristic deviation amount, and comparing the corrected characteristic deviation amount with a sensitivity threshold to filter abnormal fluctuation caused by accidental behaviors in the multi-mode academic performance data.
3. The full-flow academic employment guidance system based on the big model and the multi-modal data according to claim 1, wherein the multi-modal academic performance data comprises text of a contest training link, audio-video stream data, academic achievement text data, physiological characteristic data of psychological test and personal wish investigation text data of students.
4. The full-flow academic employment guidance system based on the large model and the multi-mode data according to claim 1, wherein the feature mapping module uses the hidden layer state of the large model as the real-time capability feature vector, and the feature mapping module is further configured to attach semantic inertia weights to the real-time capability feature vector according to the historical behavior trace of the target individual, so as to characterize the stability of the hidden capability feature in the time dimension.
5. The full-flow employment guidance system based on the big model and the multi-mode data according to claim 1, wherein the path reconstruction module realizes the quantitative characterization of the dynamic correction matrix in the decision metadata by constructing a data matrix with the industry post increment rate as a main axis characteristic, the policy support weight as an adjustment coefficient, the technical iteration rate, the present university employment data and the user expectations as attenuation factors when executing the step S1.
6. The full-flow employment guidance system based on big model and multi-modal data as set forth in claim 1, wherein the feature deviation determination module calculates the euclidean distance between the real-time capability feature vector and the preset industry reference anchor point when calculating the feature deviation amount, the feature deviation amount The calculation rule of (2) is as follows: , wherein, In the first place for the real-time capability feature vector The component values of the individual dimensions are used, Is at the first industry standard anchor point The component values of the individual dimensions are used, The total dimension number of the feature space is preset.
7. The full-flow academic employment guidance system based on big models and multi-modal data according to claim 1, wherein the path reconstruction module converts a preset industry reference anchor point into a reference vector when executing step S2, and performs matrix multiplication operation with the reference vector by using a dynamic correction matrix to predict the migration position of the industry post competence boundary in the target prediction time domain.
8. The full-flow employment guidance system based on the big model and the multi-mode data according to claim 1, further comprising a feedback control module, wherein the feedback control module is used for monitoring the behavioral response data of the target individual to the employment guidance path in real time after the employment guidance path is generated, and dynamically fine-tuning the sensitivity threshold in the characteristic deviation judging module according to the degree of agreement between the behavioral response data and the expected target.
9. The full-flow employment guidance system based on the big model and the multi-mode data according to claim 1, wherein the data acquisition module utilizes a attention mechanism to identify core feature sites in the multi-mode employment performance data and performs weighted fusion on the identified plurality of core feature sites when the input feature set is generated, and the system further comprises an interactive presentation module for synchronously converting the real-time capability feature vector, the dynamic prediction anchor point and the employment guidance path into a three-dimensional topological relation diagram for dynamic display.
10. The full-flow academic employment guidance method based on the large model and the multi-mode data is characterized by comprising the following steps: Acquiring multi-modal academic performance data aiming at a target individual, and carrying out semantic sampling and feature dimension reduction on the multi-modal academic performance data to generate an input feature set consisting of dominant business keywords and implicit capability features; Inputting the input feature set into a large model to obtain a real-time capability feature vector representing a target individual, and determining coordinate data of the real-time capability feature vector in a preset feature space; calculating the characteristic deviation amount between the real-time capacity characteristic vector and a preset industry reference anchor point, and generating a trigger signal according to the comparison result of the characteristic deviation amount and a sensitivity threshold, wherein the sensitivity threshold is a dynamic variable set according to the historical prediction accuracy; The method comprises the following steps of responding to a trigger signal, executing path correction operation, namely, step S1, calling external decision metadata comprising user expectations, the employment data of the present university, industry post increment rate, policy support weight and technical iteration rate to construct a dynamic correction matrix representing industry development trend, step S2, utilizing the dynamic correction matrix as a linear predictor to conduct weighted offset calculation on a preset industry reference anchor point to obtain a dynamic prediction anchor point aligned with a target prediction time domain, and step S3, regenerating an employment guidance path for a target individual according to spatial distribution residual errors between a real-time capability feature vector and the dynamic prediction anchor point.

Description

Full-flow academic employment guidance system and method based on large model and multi-mode data Technical Field The invention belongs to the technical field of application of artificial intelligence in education management, and particularly relates to a full-flow academic employment guidance system and method based on a large model and multi-mode data. Background Currently, an artificial intelligence technology generally adopts a static tag matching mode in the fields of education decision-making and prediction management, dominant keywords in resume data are extracted, a preset post competence library is compared, and the mode realizes primary screening of talent supply and demand by establishing a standardized evaluation system, so that references are provided for overall management decisions. However, as the adjustment speed of the industrial structure increases, time domain deviation is generated between the education culture period and market demand change, the existing scheme mainly focuses on the static attribute of an individual, ignores the dynamic evolution characteristic in the process of learning growth, cannot extract the recessive capability characteristic in multi-mode behavior data, generates prediction hysteresis when processing long-period education output, when the system issues a path correction instruction, the original post skill boundary shifts, and causes management failure, for example, the utility model patent with the authorized bulletin number of CN205486963U discloses employment guiding equipment, acquires the employment intention form and identity information of the student by utilizing a terminal and collects the information to a remote server, the mode focuses on data transfer, cannot quantitatively evaluate the individual capability evolution trend, lacks sensitivity capture of industry increment, policy guiding and technical iteration rate, causes phase deviation between the guiding instruction and market demand, and attempts to select paths such as increasing the calculation frequency of the whole data or adopting a generated reasoning model, and the like, so that the improvement path not only increases the calculation load of the system, but also lacks recognition means of behavior motivation the deviation, generates oscillation when the individual is faced with abnormal, and finally generates the deviation of phase fluctuation, thus the education resource cannot be caused. The invention is based on a three-dimensional analysis framework of subjective expectation of student individual academic and professional development, dynamic demand of market industry and historical employment data of institutions, and aims to construct an intelligent decision support system oriented to student growth process so as to realize the recommendation of academic planning and employment guidance schemes which are matched with the development state of students in real time, How to realize monitoring capability feature drift, restrain behavior noise interference and dynamically compensate a target anchor point becomes the technical problem to be solved by the invention. Disclosure of Invention The invention provides a full-flow academic employment guidance system based on large models and multi-mode data, which comprises a data acquisition module, a feature mapping module, a feature deviation judging module and a path reconstruction module, wherein the data acquisition module is used for acquiring the data of the full-flow academic employment guidance system based on the large models and the multi-mode data: The data acquisition module is used for acquiring multi-modal academic performance data aiming at a target individual, and carrying out semantic sampling and feature dimension reduction on the multi-modal academic performance data so as to generate an input feature set consisting of dominant business keywords and implicit capability features; The feature mapping module is used for inputting the dominant business keywords and the recessive capability features into the large model to obtain real-time capability feature vectors representing target individuals, and determining coordinate data of the real-time capability feature vectors in a preset feature space; The characteristic deviation judging module is used for calculating the characteristic deviation between the real-time capacity characteristic vector and a preset industry reference anchor point and generating a trigger signal according to the comparison result of the characteristic deviation and a sensitivity threshold, wherein the sensitivity threshold is a dynamic variable set according to the historical prediction accuracy; The path reconstruction module is used for responding to a trigger signal to execute the following path correction steps of S1, calling decision metadata comprising user expectations, the employment data of the present university, industry post increment rate, policy support weight, technical iteration rate a