Large model driven resource-demand integrated evaluation system

Abstract

The application belongs to the technical field of business processing, and discloses a resource-demand integrated evaluation system driven by a large model; comprises the following steps: collecting multi-source business data of a school-enterprise cooperation process and performing data cleaning to obtain standard business data; performing semantic analysis and performing term specification processing to obtain a domain term table; performing semantic coding to obtain an option feature vector and constructing a combined feature vector set; performing dimension classification and constructing a feature alignment matrix, and simultaneously constructing a matching feature space; obtaining dimension weights and performing matching fusion, and outputting a response processing sequence for demand calling; extracting project effectiveness information and performing semantic labeling; performing aggregated calculation on the project effectiveness information, and outputting an objective index score; constructing an effectiveness feedback record table, and simultaneously performing evaluation process parameter updating; and the integrated level of school-enterprise cooperation resource matching and evaluation processing is improved.

Description

Technical Field

[0001] This invention relates to the field of business processing technology, and more specifically, to a large-model-driven integrated resource-demand evaluation system. Background Technology

[0002] Resource allocation, demand matching, and effectiveness evaluation in school-enterprise cooperation are crucial management aspects for vocational schools in promoting industry-education integration and collaborative talent cultivation. With the continuous growth in the number of cooperating units, cooperation formats, and project outcome types, traditional management methods relying on manual screening, classification, and data entry are no longer sufficient to meet the data processing needs of complex school-enterprise cooperation scenarios. Therefore, utilizing large-scale models to process various types of data in the school-enterprise cooperation process, and on this basis, achieving integrated resource-demand evaluation, has become an important technological direction for improving the digitalization and intelligence of school-enterprise cooperation management. However, traditional evaluation systems still face many challenges.

[0003] In real-world scenarios, both academic profiles and collaborating unit profiles often contain structured fields, free text fields, and multiple-choice fields. Different units may use abbreviations, aliases, and non-standard expressions to describe specialized concepts in areas such as industry-university collaboration, construction of two-body systems, field engineer training programs, industry professors, and hardware construction. Traditional evaluation systems typically rely solely on fixed-field comparisons or keyword searches for identification, lacking the ability to perform semantic completion and terminology standardization based on large-scale models. This leads to the inaccurate merging of synonymous concepts in resource profiles and requirement descriptions, affecting subsequent matching results. Furthermore, there is a significant dimensional asymmetry between academic and collaborating unit profiles. For example, the academic side emphasizes institutional mechanisms, personnel size, and professional capabilities, while the collaborating unit side emphasizes the nature of the unit, its level, and collaborative capabilities. Traditional evaluation systems often assume symmetrical field structures for both sides, only addressing a small number of related aspects. Simple comparisons of the same field cannot establish a unified representation for asymmetric dimensions. Furthermore, different types of collaborations focus on different aspects of matching dimensions. For example, technical breakthroughs emphasize technical direction, while order-based training and on-site engineer training emphasize the form of collaboration and historical collaboration. However, existing technologies typically use fixed weights to calculate comprehensive matching scores, which is insufficient to meet the resource recommendation needs of different collaboration types. On the other hand, the data types of results generated in project management are complex and their field structures vary significantly. Evaluation systems often employ multi-level tree-like indicator structures. Traditional evaluation systems still rely heavily on manual categorization of project results into specific indicators and manual score calculation. They lack the ability to automatically map result fields to evaluation indicator nodes using large models and terminology systems, making it difficult to generate accurate resource-demand integrated evaluation results in complex business scenarios for university-enterprise collaborations.

[0004] In view of this, the present invention proposes a large-scale model-driven integrated resource-demand evaluation system to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: a large-model-driven integrated resource-demand evaluation system, characterized in that it includes: The data acquisition module collects multi-source business data from the school-enterprise cooperation process and performs data cleaning to obtain standard business data; the standard business data includes profile data, process data, and indicator data. The semantic parsing module performs semantic parsing on text-type data in standard business data, performs terminology standardization processing based on the semantic parsing results, and maps the content of different expressions in a unified manner to obtain a domain terminology table. The feature combination module extracts multiple-choice fields from the profile data, performs semantic encoding on the options of each multiple-choice field based on the domain terminology to obtain option feature vectors, calculates the correlation between options, and constructs a set of combined feature vectors based on the correlation and option feature vectors. The spatial construction module performs dimensional classification on the portrait data based on the combined feature vector set, constructs a feature alignment matrix based on the classification results, and builds a matching feature space based on the feature alignment matrix. The weight adjustment module extracts the cooperation content from the process data and obtains the dimension weights from the preset weight rule library based on the cooperation content; it performs matching and fusion based on the dimension weights in the matching feature space and outputs a response processing sequence for request invocation; The performance evaluation module extracts project performance information from process data, performs semantic annotation on the project performance information based on a domain terminology list, performs location matching and establishes a mapping relationship between the annotation results and indicator data, and aggregates and calculates the project performance information based on the mapping relationship to output objective indicator scores. The report generation module associates objective indicator scores with corresponding collaborative content and response processing sequences to form an effectiveness feedback record table, and updates the evaluation process parameters based on the effectiveness feedback record table; the various modules are connected via wired and / or wireless means.

[0006] Furthermore, the method for performing semantic parsing includes: Text-type data is used as a set of text to be parsed. This set of text to be parsed is then input into a large model to perform word segmentation to obtain a word sequence. Traverse the lexical sequence to identify noun lexical units, extract noun lexical units and lexical combinations of adjacent noun lexical units to generate a candidate term lexical unit set; The candidate terminology set is matched with the preset school-enterprise domain terminology database, and the candidate terminology that can completely match the standard terminology is marked as the direct recognition result. For candidate term elements that do not fully match, extract the context fragments of a preset number of elements before and after their position, input the context fragments into the large model to identify the semantic proximity relationship between the candidate term element and each standard term, and output the identification confidence score. Candidate terminology units with confidence scores higher than the confidence threshold are selected and identified, and their corresponding standard terms are marked as indirect identification results. The direct and indirect identification results are then integrated to obtain the semantic parsing results.

[0007] Furthermore, the method of performing terminology standardization processing includes: Read the confirmed term elements that have been matched with standard terms in the semantic parsing results, and group the confirmed term elements that correspond to the same standard terms into a set of aliases for the corresponding standard terms; Count the frequency of each standard term in text data, and mark standard terms that appear more frequently than a preset frequency threshold as high-frequency terms; The terminology category and level of each standard term are obtained from the pre-set terminology database of the school-enterprise field. The terminology category, terminology level and alias set are associated to generate the standard terminology specification record. The standard records of corresponding standard terms are grouped according to term categories, and then sorted according to term hierarchy within each group to form a domain terminology table. The alias set of each standard term in the domain terminology table is compared with the original text content in the text type data, and the aliases of each standard term in the original text are replaced with the corresponding standard terms.

[0008] Furthermore, the method of performing semantic encoding includes: The multiple-select type fields include the technical direction field and cooperation content field for college profiles and partner unit profiles, which are merged into a set of multiple-select fields; Identify each multiple-choice field in the set of multiple-choice fields, and read the content of each option in each multiple-choice field; match the content of each option with the standard terms in the domain terminology table to confirm the standard terms, term categories and term levels corresponding to each option content; Using standard terms as an index, extract the semantic vectors of the terms corresponding to each option from the preset term vector database; convert term categories into category vectors and term levels into level vectors; The terminology semantic vector, category vector, and hierarchy vector are concatenated to obtain the option feature vector of each option content; Sort the option feature vectors of each option within the same multi-select field to obtain the option feature vector sequence corresponding to each multi-select field.

[0009] Furthermore, the method for constructing the combined feature vector set includes: Calculate the vector similarity between any two option feature vectors in the option feature vector sequence corresponding to the same multi-select field, and construct the option similarity matrix; Normalize the vector similarity corresponding to each element in the option similarity matrix to obtain the option correlation degree between each option pair; A correlation weight matrix is ​​constructed based on the correlation between each pair of options. The correlation weight matrix is ​​then fused with the sequence of option feature vectors to obtain the weighted feature vector of each option content. The number of options in each multiple-choice field is counted, and an option number normalization factor is calculated based on the number of options. The weighted feature vectors of each option are summed, and the product of this sum and the option number normalization factor is calculated to obtain the combined feature vector of the corresponding multiple-choice field. The combined feature vectors of all multiple-choice fields are integrated to form a set of combined feature vectors.

[0010] Furthermore, the method for constructing the feature alignment matrix includes: The classification results include symmetric dimension labels and asymmetric dimension labels. Symmetric dimension labels refer to dimension labels that are the same in both the college profile and the partner unit profile in the profile data. Asymmetric dimension labels refer to dimension labels that exist only in the college profile or the partner unit profile. Obtain the combined feature vector corresponding to each symmetric dimension identifier, and pair the combined feature vector corresponding to the symmetric dimension identifier in the college profile with the combined feature vector corresponding to the collaborating unit profile to obtain the common dimension feature pair; Identify the intermediate dimension identifier corresponding to each asymmetric dimension identifier, and obtain the intermediate semantic vector corresponding to the intermediate dimension identifier. Concatenate the combined feature vector corresponding to the asymmetric dimension identifier with the intermediate semantic vector to obtain the compensation feature vector and add a data source label. Construct a common dimension matrix by arranging all common dimension feature pairs in the order of dimension labels, and construct an asymmetric dimension matrix by arranging each compensation feature vector in the order of dimension labels. The feature alignment matrix is ​​obtained by concatenating the common dimension matrix and the asymmetric dimension matrix; a matching feature space is constructed based on the feature alignment matrix, and the college profile and the partner unit profile are mapped to the matching feature space.

[0011] Furthermore, the matching and fusion methods include: Identify the collaboration type tags and the effective deadline of the requirements for the collaboration content, and simultaneously call the corresponding dimension weight settings based on the collaboration content type tags; Calculate the time interval between the effective deadline of the demand and the latest collection date, and construct an urgency adjustment coefficient based on the time interval; extract the geographic location weight in the dimension weight setting, and calculate the product of the urgency adjustment coefficient and the geographic location weight to obtain the adjusted geographic location weight; Extract the historical number of collaborations and the most recent collaboration date from the profile data, and calculate the collaboration activity coefficient based on the historical number of collaborations and the most recent collaboration date; In the matching feature space, calculate the dimensional matching value between the cooperation content and each cooperating unit in the dimensions of technology field, technology direction, geographical location and cooperation type. The matching value of each dimension is weighted and fused with the corresponding dimension weight, and the comprehensive matching score of each cooperating unit is obtained by combining the cooperation activity coefficient. The collaborating units are sorted based on their comprehensive matching scores, and those with comprehensive matching scores higher than a preset matching score threshold are selected and arranged into a response processing sequence.

[0012] Furthermore, the method for performing location matching and establishing mapping relationships includes: Identify the outcome type tags and corresponding outcome field sets of project outcome information, and extract the field names and field contents of the outcome fields; match the field names and field contents with the domain terminology table to determine the standard terms and term categories of the corresponding outcome fields; Construct an evaluation indicator tree based on indicator data, traverse all indicator nodes in the evaluation indicator tree, and read the indicator name, indicator level and related terms of each indicator node. The standard terms of each result field are compared with the indicator-related terms of each indicator node. The indicator nodes that have an intersection between the standard terms and the indicator-related terms are identified as candidate mapping indicator nodes for the corresponding result fields. Based on the indicator hierarchy, candidate mapping indicator nodes for all outcome fields are divided, and the indicator node with the deepest indicator hierarchy is retained as the target mapping indicator node; the mapping relationship between each outcome field and the corresponding target mapping indicator node is established.

[0013] Furthermore, the method for performing aggregation calculations includes: Based on the mapping relationship and target mapping indicator nodes, all result field sets are grouped to obtain the result field group corresponding to each indicator node; the result field group corresponding to each indicator node is traversed to identify the field type of each result field, including numeric type, option type and text type; Extract the field values ​​of numerical result fields and perform mathematical operations based on preset aggregation rules; extract the field options of option type result fields and convert the field options into corresponding option scores; identify the field content of text type result fields, determine the number of standard terms in the field content based on the domain terminology table, and use the number of standard terms as the text score; The mathematical operation results, option scores, and text scores corresponding to the same indicator node are summed to generate the original indicator score for that indicator node. The original indicator scores of the indicator nodes are normalized to obtain the objective indicator scores. At the same time, the objective indicator scores of all indicator nodes are rewritten into the corresponding indicator nodes in the evaluation indicator tree for updating.

[0014] Furthermore, the methods for updating the parameters of the performance evaluation process include: Based on the statistical analysis of objective indicator scores and response processing sequence ranking results corresponding to different cooperation contents in the performance feedback record table, the evaluation parameter correction amount is constructed and written into the preset evaluation standard parameter table for updating.

[0015] The technical effects and advantages of the large-model-driven integrated resource-demand evaluation system of this invention are as follows: By collecting multi-source business datasets from the university-enterprise cooperation process in real time, and performing semantic parsing, semantic encoding, profile dimension alignment, and evaluation indicator mapping analysis on this dataset, a large-model-driven integrated resource-demand evaluation system was constructed. Compared with existing technologies, the system improves the accuracy of identifying specialized terms in the university-enterprise cooperation field and normalizing the mapping of different expression methods by performing unified semantic parsing on various texts. By performing semantic encoding and correlation fusion on multiple options in the technical direction field and cooperation content field, the system optimizes the problem of combined semantic distortion. By performing pairing alignment and intermediate semantic compensation on the symmetrical and asymmetrical dimensions in the college profile and the cooperation unit profile respectively, the system avoids the failure of the matching model caused by the inconsistency of the field structure of the dual profile system. By calling differentiated dimension weights according to the cooperation content type and combining them with historical cooperation activity to perform matching fusion, the system adjusts the adaptation relationship between resource recommendation ranking and actual needs. By automatically matching and performing aggregation calculation on the heterogeneous project effectiveness field with the evaluation indicator tree, the system realizes the automatic generation of objective indicator scores and the closed-loop update of evaluation parameters, improving the integration level of university-enterprise cooperation resource matching and evaluation processing. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the resource-demand integrated evaluation system driven by the large model of the present invention. Detailed Implementation

[0017] 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.

[0018] Example 1 Please see Figure 1 As shown in this embodiment, the large-model-driven integrated resource-demand evaluation system includes: The data acquisition module collects multi-source business data from the school-enterprise cooperation process and performs data cleaning to obtain standard business data; the standard business data includes profile data, process data, and indicator data. The semantic parsing module performs semantic parsing on text-type data in standard business data, performs terminology standardization processing based on the semantic parsing results, and maps the content of different expressions in a unified manner to obtain a domain terminology table. The feature combination module extracts multiple-choice fields from the profile data, performs semantic encoding on the options of each multiple-choice field based on the domain terminology to obtain option feature vectors, calculates the correlation between options, and constructs a set of combined feature vectors based on the correlation and option feature vectors. The spatial construction module performs dimensional classification on the portrait data based on the combined feature vector set, constructs a feature alignment matrix based on the classification results, and builds a matching feature space based on the feature alignment matrix. The weight adjustment module extracts the cooperation content from the process data and obtains the dimension weights from the preset weight rule library based on the cooperation content; it performs matching and fusion based on the dimension weights in the matching feature space and outputs a response processing sequence for request invocation; The performance evaluation module extracts project performance information from process data, performs semantic annotation on the project performance information based on a domain terminology list, performs location matching and establishes a mapping relationship between the annotation results and indicator data, and aggregates and calculates the project performance information based on the mapping relationship to output objective indicator scores. The report generation module associates objective indicator scores with corresponding collaborative content and response processing sequences to form an effectiveness feedback record table, and updates the evaluation process parameters based on the effectiveness feedback record table; the various modules are connected via wired and / or wireless means.

[0019] In this embodiment, multi-source business data is directly collected through the data acquisition terminal of the university-enterprise cooperation platform. Data cleaning of multi-source business data is achieved through filtering, noise reduction, and missing value imputation to obtain higher-quality standard business data. The profile data includes profiles generated for colleges and profiles generated for partner units. College profiles include information such as the college's field, technical direction, cooperative content that can be undertaken, college introduction, personnel and management organization introduction, and college management system. Partner unit profiles refer to company-related attribute information such as unit type, unit nature, technical direction, cooperation capabilities, and project information. Process data includes information reflecting the overall cooperation process, such as cooperation requirements, cooperation contract information, work plan, university-enterprise activities, and cooperation results. Indicator data refers to the evaluation standards set by the university-enterprise cooperation platform for performance evaluation.

[0020] The methods for performing semantic parsing include: Text data is used as a set of text to be parsed. This set of text to be parsed is then input into a large model to perform word segmentation processing to obtain a word sequence.

[0021] Text-type data refers to data described using text in standard business data, including information such as college introduction text, unit introduction text, and cooperation requirement text. All text-type data is used as the set of text to be parsed in this embodiment. A large model is pre-built in this embodiment, and all technical features of this embodiment are driven by the large model. The large model is a collection of multiple sub-models. For example, in the current semantic parsing process, the set of text to be parsed is selected as the input data of the semantic processing model in the large model. The semantic processing model is used to segment all the sentences in the text to be parsed into words with independent semantics through word segmentation, and the words are sorted according to the original sentence order to form a word sequence.

[0022] Traverse the lexical sequence to identify noun lexical units, extract noun lexical units and lexical combinations of adjacent noun lexical units to generate a candidate term lexical set.

[0023] By traversing the lexical sequence, single noun terms and compound terms composed of multiple lexical units are simultaneously included in the candidate term lexical set, so that business concepts appearing in the text as phrases are not broken up.

[0024] The candidate terminology set is matched with a pre-set terminology database for the industry and academia, and candidate terminology that can completely match the standard terminology is marked as the direct recognition result.

[0025] The terminology database for the industry-university cooperation field refers to a terminology database based on the standards for industry-university cooperation. The database stores technical terms specifically for the corresponding industry-university cooperation field. The candidate terminology set is matched with the terminology database for the industry-university cooperation field. Terms that have been filled in according to the standard format in the text are identified first, so that they can be quickly matched with standard terms and used as direct identification results, reducing the processing scope of subsequent operations.

[0026] For candidate term elements that do not fully match, extract the context fragments of a preset number of elements before and after their position, input the context fragments into the large model to identify the semantic proximity relationship between the candidate term element and each standard term, and output the identification confidence score.

[0027] Among them, there are candidate terminology units that cannot be directly matched with standard terms. It is difficult to determine whether they belong to the proprietary concepts of the corresponding school-enterprise cooperation field based solely on their literal form. Therefore, a preset number of context segments are extracted before and after the terminology. The preset number of terminology units is set based on historical terminology matching experience to ensure that the range can include the longest term length under normal circumstances. The semantic processing model is used in combination with the restrictive semantics of the context segments to determine the semantic relationship between the candidate terminology units that do not fully match and each other standard term. Based on the semantics of the context segments, semantic completion is performed on abbreviations, contractions, colloquial expressions and policy terms, and the recognition confidence corresponding to the semantic proximity relationship is output.

[0028] Candidate terminology units with confidence levels higher than the confidence threshold are selected and identified, and their corresponding standard terms are marked as indirect identification results.

[0029] The parameter configuration based on the semantic processing model sets the confidence threshold in the semantic parsing scenario and selects candidate terminology units that meet the target requirements in terms of semantic association with standard terms as indirect recognition results.

[0030] The semantic parsing result is obtained by integrating the direct recognition result and the indirect recognition result of the large model-assisted recognition.

[0031] The methods for implementing terminology standardization include:

[0032] Read the confirmed term tokens that have been matched with standard terms in the semantic parsing results, and group the confirmed term tokens that correspond to the same standard terms into a set of aliases for the corresponding standard terms.

[0033] In this process, lexical units that can match standard terms in the semantic parsing results are identified as lexical units, while lexical units of multiple different spellings but semantically belonging to the same standard term are included in the alias set of the corresponding standard term.

[0034] Count the frequency of each standard term in text data, and mark the standard terms that appear more frequently than a preset frequency threshold as high-frequency terms.

[0035] The semantic recognition standard based on the semantic processing model sets a frequency threshold. When the frequency of a certain standard term in text data is higher than the preset frequency threshold, the standard term is marked as a high-frequency term.

[0036] The terminology category and level of each standard term are obtained from the pre-set terminology database of the school-enterprise field. The terminology category, terminology level and alias set are associated to generate the standard terminology specification record.

[0037] The terminology database for the industry-university sector not only stores all standard terms, but also records the attribute information of the corresponding standard terms, including term category and term level. It combines and associates term categories, term levels and aliases belonging to the same standard term to form the normative record of the corresponding standard term.

[0038] The standard records of corresponding standard terms are grouped according to term categories, and then sorted according to term hierarchy within each group to form a domain terminology table. The domain terminology table, which is grouped by term category and sorted according to term hierarchy, has clear boundaries, clear hierarchical relationships and complete alias information.

[0039] The alias set of each standard term in the domain terminology table is compared with the original text content in the text type data. The aliases of each standard term in the original text are replaced with the corresponding standard terms. The domain terminology table is used as the basis to compare and replace the aliases of each standard term in the original text item by item, so that all texts are unified into a standardized expression.

[0040] Methods of semantic encoding include: The multiple-choice fields include the technical direction field and the cooperation content field for both the college profile and the partner unit profile, which are then merged into a multiple-choice field set.

[0041] The "Technology Direction" field is used to characterize the technical capabilities of the college or partner institution in the industry-university cooperation, while the "Cooperation Content" field is used to reflect the forms and content of cooperation that the college or partner institution can participate in.

[0042] Identify each multi-select field in the set of multi-select fields, and read the content of each option in each multi-select field. The content of each option in each multi-select field refers to the discrete option content of the technology direction and cooperation content filled in in the form of multiple selections, such as fields such as intelligent manufacturing or industrial robots.

[0043] Match each option with the standard terms in the domain terminology table to confirm the standard terms, term categories, and term levels corresponding to each option.

[0044] This involves matching the content of each option in each multiple-choice field with the domain terminology table to determine the standard terminology, terminology category, and terminology level of the corresponding option. The terminology category is used to distinguish the technical direction category or cooperation content category of the corresponding option, and the terminology level is used to indicate the hierarchical relationship of the corresponding option in the entire school-enterprise cooperation business process.

[0045] Using standard terms as an index, extract the semantic vectors of the terms corresponding to each option from a pre-defined term vector database. The term vector database stores the vector representations of standard terms processed by a language processing model, and extracts the semantic vectors of the standard terms corresponding to each option.

[0046] The term categories are converted into category vectors, and the term levels are converted into level vectors. The term categories and term levels are converted into vector representations to obtain the category vectors and level vectors of the corresponding standard terms.

[0047] The term semantic vector, category vector, and hierarchy vector are concatenated to obtain the option feature vector of each option content. Specifically, the option feature vector of the corresponding option is obtained by concatenating the term semantic vector corresponding to the standard term of each option content with the category vector and hierarchy vector, so that the vector simultaneously includes semantic information, category information, and hierarchy information.

[0048] The option feature vectors of each option within the same multi-select field are sorted to obtain the option feature vector sequence corresponding to each multi-select field. The sorting method refers to sorting the option feature vectors of each option within the same multi-select field according to the terminology level of the corresponding standard terminology.

[0049] Methods for constructing a set of combined feature vectors include: Calculate the vector similarity between any two option feature vectors in the option feature vector sequence corresponding to the same multi-select field, and construct the option similarity matrix.

[0050] The method calculates vector similarity using cosine similarity and uses the vector similarity as matrix elements to represent the degree of similarity between any pair of options. This is used to distinguish whether the contents of each option in a multiple-choice field are highly related, partially related, or weakly related. For example, in the technology field, intelligent manufacturing and industrial robots may have a high degree of similarity, while industrial robots and biomedicine may have a low degree of similarity.

[0051] Normalize the vector similarity corresponding to each element in the option similarity matrix to obtain the option correlation degree between each option pair.

[0052] Since the data sources and distributions of the feature vectors of different options in different multiple-choice fields may be different, directly using the original vector similarity will lead to the inability to directly compare the similarity relationships between different multiple-choice fields. Therefore, normalization processing is performed on each element in the option similarity matrix to uniformly map the similarity between each pair of options to a comparable numerical range, thereby obtaining the option correlation degree and enhancing the comparability of the combined features of multiple-choice fields between different portrait subjects.

[0053] A correlation weight matrix is ​​constructed based on the correlation between each pair of options. The correlation weight matrix is ​​then fused with the sequence of option feature vectors to obtain the weighted feature vector of each option content.

[0054] The correlation weight matrix is ​​used to represent the semantic correlation strength between the contents of each option in the same multi-choice field, and is explicitly represented by a weight structure. The corresponding option feature vectors in the option feature vector sequence are weighted by the correlation of all options in each row of the correlation weight matrix and summed to obtain the weighted feature vector corresponding to each option content.

[0055] The number of options in each multi-select field is counted, and the option number normalization factor is calculated based on the number of options. The reciprocal of the option number is used as the option number normalization factor. The option number refers to the number of discrete option contents in the multi-select field.

[0056] The weighted feature vectors of each option are summed, and the product of this sum and the option number normalization factor is calculated to obtain the combined feature vector of the corresponding multi-select field. The combined feature vectors of all multi-select fields are integrated to form a set of combined feature vectors.

[0057] The formula for calculating the combined feature vector is as follows: ;in, This represents a combined feature vector of any multiple-selection field. Indicates the number of options; Indicates the first option in the corresponding multi-select field. The weighted feature vectors of each option's content enable the combined feature vectors to simultaneously reflect the semantic content of each option within a multi-select field, the relationships between options, and the differences in the number of options.

[0058] Methods for constructing feature alignment matrices include: The classification results include symmetric dimension labels and asymmetric dimension labels. Symmetric dimension labels refer to dimension labels that are the same in both the college profile and the partner unit profile in the profile data. Asymmetric dimension labels refer to dimension labels that exist only in the college profile or the partner unit profile.

[0059] The dimensions present in the two types of profiles, namely the college profile and the partner profile, are divided into symmetrical dimensions and asymmetrical dimensions. Symmetrical dimensions include fields related to the technical field, technical direction, and cooperation content that are common to both types of profiles. Asymmetrical dimensions include fields that exist only in the college profile, such as fields related to the college system and the size of the college personnel, or fields that exist only in the partner profile, such as fields related to the type of organization and the nature of the organization.

[0060] Obtain the combined feature vector corresponding to each symmetric dimension identifier, and pair the combined feature vector corresponding to the symmetric dimension identifier in the college profile with the combined feature vector corresponding to the collaborating unit profile to obtain the shared dimension feature pair.

[0061] Since the combined feature vector of the multi-select field is a unified feature representation of the field-level dimension, and the symmetric dimension identifier is also a representation method of the field-level dimension, the two combined feature vectors corresponding to the symmetric dimension identifiers shared by the college profile and the partner unit profile can be paired to form a shared dimension feature pair, which is used to represent the correspondence between the college and the partner unit in the same business dimension.

[0062] Identify the intermediate dimension identifier corresponding to each asymmetric dimension identifier, and obtain the intermediate semantic vector corresponding to the intermediate dimension identifier. Concatenate the combined feature vector corresponding to the asymmetric dimension identifier with the intermediate semantic vector to obtain the compensation feature vector and add the data source mark.

[0063] The intermediate dimension identifiers are set based on relevant content in the college data database and the partner unit data database. They map the profile-related fields that cannot be directly matched in the original asymmetric dimension identifiers to an intermediate semantic layer with general explanatory meaning. For example, institutional mechanism fields such as the college council and the school-enterprise cooperation organization can be mapped to intermediate dimensions such as organizational guarantee capability or institutional guarantee capability. The intermediate semantic vectors are also extracted from the pre-set term vector database. Based on the combined feature vectors corresponding to the asymmetric dimension identifiers, intermediate semantic vectors are introduced to bridge information and obtain a compensating feature vector. At the same time, a data source mark is added to identify whether the compensating feature vector comes from the college profile or the partner unit profile.

[0064] Construct a common dimension matrix by arranging all common dimension feature pairs in the order of dimension labels, and construct an asymmetric dimension matrix by arranging each compensation feature vector in the order of dimension labels.

[0065] This involves combining and arranging common dimensional features into a matrix form according to a preset dimensional label order, and also arranging all compensation feature vectors into a matrix form according to a preset dimensional label order. The dimensional label order refers to the order of field importance set in the cooperation plan.

[0066] The feature alignment matrix is ​​obtained by concatenating the common dimension matrix and the asymmetric dimension matrix.

[0067] The feature alignment matrix is ​​obtained by concatenating the common dimension matrix and the asymmetric dimension matrix. If there is a mismatch between rows and columns, zeros are used to fill in the rows and columns to ensure that the two matrices can be concatenated.

[0068] A matching feature space is constructed based on the feature alignment matrix, and the college profile and the profile of the cooperating unit are mapped to this matching feature space.

[0069] The feature alignment matrix is ​​used as the input basis for unified features to construct a matching feature space that simultaneously accommodates shared dimension features and asymmetric dimension compensation features. This allows the college profile and the profile of the cooperating unit to be represented in the matching feature space as profile feature representations with consistent structure and comparable dimensions.

[0070] The methods for matching and fusion include: Identify the collaboration type tags and the effective deadline of the requirements for the collaboration content, and simultaneously call the corresponding dimension weight settings based on the collaboration content type tags.

[0071] The collaboration content type tags include, for example, course development, practical training, or technical problem-solving. Based on the collaboration type tags, the corresponding weight settings of various dimensions are identified from the weight rule library set based on the collaboration plan content, so that different collaboration content types correspond to different weight configurations, including technical field dimension, technical direction dimension, geographical location dimension, and collaboration type dimension; the effective deadline of the requirement refers to the time range limit set for the corresponding collaboration plan.

[0072] Calculate the time interval between the effective deadline of the demand and the latest collection date, and construct an urgency adjustment coefficient based on the time interval.

[0073] The formula for calculating the stress adjustment coefficient is as follows: ;in Indicates the urgency adjustment coefficient; Indicates a time interval; This represents the time reference constant set based on historical coefficients and empirically constructed.

[0074] Extract the geographic location weight from the dimension weight settings, and calculate the product of the urgency adjustment coefficient and the geographic location weight to obtain the adjusted geographic location weight.

[0075] The geographical location weight in the dimension weight setting is extracted and weighted using an urgency adjustment coefficient. This makes the proximity between the candidate partner and the location of the required partner account for a higher proportion in the matching when the processing time is tight, thereby improving the feasibility of completing the processing of urgent needs within the time limit.

[0076] Extract the historical number of collaborations and the most recent collaboration date from the profile data, and calculate the collaboration activity coefficient based on the historical number of collaborations and the most recent collaboration date.

[0077] The collaboration activity coefficient is obtained by normalizing the number of historical collaborations and the time interval between the current time and the most recent collaboration time, and then multiplying the normalized values. The number of historical collaborations is used to characterize the degree of accumulated collaboration experience, and the most recent collaboration date is used to characterize the timeliness of the collaboration relationship.

[0078] In the matching feature space, calculate the dimensional matching value between the cooperation content and each cooperating unit in the dimensions of technology field, technology direction, geographical location and cooperation type.

[0079] This involves constructing a semantic vector of the collaborative content in the matching feature space, and calculating the cosine similarity between this semantic vector and the vector representation of each collaborative unit in each dimension of the dimension weight setting. The resulting cosine similarity is the dimension matching value between the collaborative content and the corresponding collaborative unit in each dimension.

[0080] The matching value of each dimension is weighted and fused with the corresponding dimension weight, and the comprehensive matching score of each cooperating unit is obtained by combining the cooperation activity coefficient.

[0081] The matching values ​​for each dimension are weighted using the corresponding dimension weights and summed. The sum is then weighted using the cooperation activity coefficient to obtain the comprehensive matching score for the corresponding cooperating unit. The geographical location weight is an adjusted geographical location weight.

[0082] The collaborating units are sorted based on their comprehensive matching scores, and those with comprehensive matching scores higher than a preset matching score threshold are selected and arranged into a response processing sequence.

[0083] The system sets a matching score threshold based on the demand response requirements set in the cooperation plan. Candidates with low matching scores are excluded, and the cooperation units that meet the minimum adaptation requirements are arranged into a response processing sequence according to their comprehensive matching scores, which is used for subsequent demand processing calls.

[0084] The methods for performing location matching and establishing mapping relationships include: Identify the outcome type tags and corresponding outcome field sets for project outcome information, and extract the field names and field contents of the outcome fields.

[0085] The achievement type tags include, for example, construction, industry, and technological breakthroughs. After determining the achievement type tag, the achievement fields used to describe the achievement are identified. The field name refers to the name of each specific project in the project achievement information. For example, in the achievement of textbook construction cooperation, the fields such as textbook name or textbook level are field names, and the subsequent specific descriptions of the textbook name or specific textbook level are field content.

[0086] Match the field names and field contents with the domain terminology table to determine the standard terms and term categories for the corresponding output fields.

[0087] This involves matching field names and content with a domain terminology table, mapping different information to a unified terminology system, and determining the standard terminology corresponding to the output field and the terminology category to which the standard terminology belongs.

[0088] Construct an evaluation indicator tree based on indicator data, traverse all indicator nodes in the evaluation indicator tree, and read the indicator name, indicator level, and related terms of each indicator node.

[0089] The indicator data includes the indicator name, indicator level, parent indicator identifier, indicator type, and related terms. By reading the indicator level and parent indicator identifier from each indicator data, the first-level indicator node without a parent indicator identifier is determined as the root node of the tree. Then, the second-level indicator nodes whose parent indicator identifier matches the existing node identifier in the tree are attached to the corresponding first-level indicator node, the third-level indicator nodes are attached to the corresponding second-level indicator nodes, and the fourth-level indicator nodes are attached to the corresponding third-level indicator nodes, thus forming a four-level tree structure of the evaluation indicator tree.

[0090] The standard terms of each outcome field are compared with the indicator-related terms of each indicator node. Indicator nodes where the standard terms and indicator-related terms intersect are identified and used as candidate mapping indicator nodes for the corresponding outcome fields.

[0091] The process involves identifying indicator nodes that correspond to each outcome field in terms of semantics through intersection comparison, and using these nodes as candidate mapping indicator nodes to narrow down the outcome fields to be judged to a range of semantically related indicator nodes.

[0092] Based on the indicator hierarchy, candidate mapping indicator nodes for all outcome fields are divided, and the indicator node with the deepest indicator hierarchy is retained as the target mapping indicator node.

[0093] The indicator level refers to the depth of the indicator node in the evaluation indicator tree. The indicator node with the deepest depth among all candidate mapping indicator nodes corresponding to all result fields is selected as the target mapping indicator node, thus obtaining the indicator node with the deepest level and the most specific semantics.

[0094] Establish the mapping relationship between each result field and the corresponding target mapping indicator node. This involves establishing a correspondence between each result field and the corresponding target mapping indicator node to form a structured mapping record from result fields to indicator nodes.

[0095] Methods for performing aggregation calculations include: Based on the mapping relationship and the target mapping indicator nodes, all result field sets are grouped to obtain the result field group corresponding to each indicator node.

[0096] The target mapping indicator node is used as the grouping basis to group and classify all result fields. Result fields that may come from different result types but ultimately belong to the same indicator node are included in the same result field group.

[0097] Iterate through the result field groups corresponding to each indicator node and identify the field type of each result field, including numeric, option, and text types.

[0098] Numerical types include, for example, the number of students covered, the number of jobs created, and the amount of funding invested; option types include, for example, information corresponding to the field level such as textbook level and the technical field corresponding to the technological achievement; text types include, for example, content expressed in text such as an overview of the cooperation content, example descriptions, and results descriptions.

[0099] Extract the numerical values ​​of the result fields of the numerical type, and perform mathematical operations based on preset aggregation rules. The preset aggregation rules refer to the corresponding mathematical calculation methods used for numerical type fields of different fields or technical directions, such as summation, counting, mean, and interval mapping.

[0100] Extract the field options of the result field of the option type, and convert the field options into the corresponding option scores.

[0101] For each field option in the result field of the option type, the different option content is converted into the corresponding option score according to the mapping rules required for that type of option. For example, fields such as school level, city level or provincial level are mapped to different scores.

[0102] The content of the output fields that identify the text type is determined, and the number of standard terms in the field content is determined based on the domain terminology table. This number of standard terms is then used as the text score.

[0103] By counting the number of standard terms in the statistical fields, descriptive text that was originally difficult to quantify is transformed into a scoreable text score, enabling narrative content in project performance information to participate in the calculation of objective indicators.

[0104] The mathematical operation results, option scores, and text scores corresponding to the same indicator node are summed to generate the original indicator score for that indicator node. The original indicator scores of the indicator nodes are normalized to obtain the objective indicator scores. At the same time, the objective indicator scores of all indicator nodes are rewritten into the corresponding indicator nodes in the evaluation indicator tree for updating.

[0105] By accumulating the scores or specific results of each type of outcome field and normalizing them, the original score deviations caused by differences in the number of fields, field distribution, and scoring scale in different indicator nodes are uniformly converted into a comparable scoring range, resulting in standardized objective indicator scores. At the same time, the obtained objective indicator scores are used to update the scores of the corresponding nodes in the evaluation indicator tree, making the evaluation indicator tree a dynamic evaluation structure that includes real-time objective scores.

[0106] The methods for updating performance evaluation process parameters include:

[0107] Based on the statistical analysis of objective indicator scores and response processing sequence ranking results corresponding to different cooperation contents in the performance feedback record table, the evaluation parameter correction amount is constructed and written into the preset evaluation standard parameter table for updating.

[0108] The performance feedback record form includes tags for the type of collaborative content, scores for objective indicators, ranking in the response processing sequence, and corresponding identifiers of collaborating units. By statistically analyzing the distribution of objective indicator scores and the ranking results of response processing sequences under different types of collaborative content, deviations between the completed matching and ranking and the subsequent actual performance are identified. Based on these deviations, as well as the requirements of the collaborative content and relevant historical experience, correction amounts for evaluation parameters are constructed, including correction amounts for indicator weights, correction amounts for all relevant thresholds, and correction amounts for scoring rules. All correction amounts for evaluation parameters are updated by writing them into the evaluation standard parameter table set based on the relevant collaborative content evaluation standards, so that subsequent evaluation processes for the same or similar types of collaborative content can adaptively adjust parameter configurations based on historical performance feedback records.

[0109] This embodiment collects multi-source business datasets from the university-enterprise cooperation process in real time. Based on this dataset, it performs semantic parsing, semantic encoding, profile dimension alignment, and evaluation indicator mapping analysis, forming a large-model-driven integrated resource-demand evaluation system. Compared with existing technologies, it improves the accuracy of identifying specialized terms in the university-enterprise cooperation field and normalizing the mapping of different expression methods by performing unified semantic parsing on various texts. It optimizes the problem of combined semantic distortion by performing semantic encoding and correlation fusion on multiple options in the technical direction field and cooperation content field. It avoids the failure of the matching model caused by the inconsistency of the field structure of the dual profile system by performing pairing alignment and intermediate semantic compensation on the symmetrical and asymmetrical dimensions of the college profile and the cooperation unit profile, respectively. It adjusts the adaptation relationship between resource recommendation ranking and actual needs by calling differentiated dimension weights according to the cooperation content type and combining historical cooperation activity to perform matching fusion. By automatically matching and performing aggregation calculations on the heterogeneous project effectiveness field with the evaluation indicator tree, it realizes the automatic generation of objective indicator scores and the closed-loop update of evaluation parameters, improving the integration level of university-enterprise cooperation resource matching and evaluation processing.

[0110] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0111] All formulas in this manual are dimensionless and calculated numerically. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0112] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A resource-demand integrated evaluation system driven by a large model, characterized in that: include: The data acquisition module collects multi-source business data from the school-enterprise cooperation process and performs data cleaning to obtain standard business data; The standard business data includes profile data, process data, and indicator data; The semantic parsing module performs semantic parsing on text-type data in standard business data, performs terminology standardization processing based on the semantic parsing results, and maps the content of different expressions in a unified manner to obtain a domain terminology table. The feature combination module extracts multiple-choice fields from the profile data, performs semantic encoding on the options of each multiple-choice field based on the domain terminology to obtain option feature vectors, calculates the correlation between options, and constructs a set of combined feature vectors based on the correlation and option feature vectors. The spatial construction module performs dimensional classification on the portrait data based on the combined feature vector set, constructs a feature alignment matrix based on the classification results, and builds a matching feature space based on the feature alignment matrix. The weight adjustment module extracts the cooperation content from the process data and obtains the dimension weights from the preset weight rule library based on the cooperation content; it performs matching and fusion based on the dimension weights in the matching feature space and outputs a response processing sequence for request invocation; The effectiveness evaluation module extracts project effectiveness information from process data, performs semantic annotation on project effectiveness information based on a domain terminology list, and performs location matching and mapping relationship between the annotation results and indicator data. Based on the mapping relationship, the project effectiveness information is aggregated and calculated to output objective indicator scores; The report generation module associates objective indicator scores with corresponding collaborative content and response processing sequences to form an effectiveness feedback record table, and updates the evaluation process parameters based on the effectiveness feedback record table; the various modules are connected via wired and / or wireless means.

2. The large-model-driven integrated resource-demand evaluation system according to claim 1, characterized in that, The methods for performing semantic parsing include: Text-type data is used as a set of text to be parsed. This set of text to be parsed is then input into a large model to perform word segmentation to obtain a word sequence. Traverse the lexical sequence to identify noun lexical units, extract noun lexical units and lexical combinations of adjacent noun lexical units to generate a candidate term lexical unit set; The candidate terminology set is matched with the preset school-enterprise domain terminology database, and the candidate terminology that can completely match the standard terminology is marked as the direct recognition result. For candidate term elements that do not fully match, extract the context fragments of a preset number of elements before and after their position, input the context fragments into the large model to identify the semantic proximity relationship between the candidate term element and each standard term, and output the identification confidence score. Candidate terminology units with confidence scores higher than the confidence threshold are selected and identified, and their corresponding standard terms are marked as indirect identification results. The direct and indirect identification results are then integrated to obtain the semantic parsing results.

3. The large-model-driven integrated resource-demand evaluation system according to claim 2, characterized in that, The methods for standardizing execution terminology include: Read the confirmed term elements that have been matched with standard terms in the semantic parsing results, and group the confirmed term elements that correspond to the same standard terms into a set of aliases for the corresponding standard terms; Count the frequency of each standard term in text data, and mark standard terms that appear more frequently than a preset frequency threshold as high-frequency terms; The terminology category and level of each standard term are obtained from the pre-set terminology database of the school-enterprise field. The terminology category, terminology level and alias set are associated to generate the standard terminology specification record. The standard records of corresponding standard terms are grouped according to term categories, and then sorted according to term hierarchy within each group to form a domain terminology table. The alias set of each standard term in the domain terminology table is compared with the original text content in the text type data, and the aliases of each standard term in the original text are replaced with the corresponding standard terms.

4. The large-model-driven integrated resource-demand evaluation system according to claim 3, characterized in that, The methods for performing semantic encoding include: The multiple-select type fields include the technical direction field and cooperation content field for college profiles and partner unit profiles, which are merged into a set of multiple-select fields; Identify each multiple-choice field in the set of multiple-choice fields, and read the content of each option in each multiple-choice field; match the content of each option with the standard terms in the domain terminology table to confirm the standard terms, term categories and term levels corresponding to each option content; Using standard terms as an index, extract the semantic vectors of the terms corresponding to each option from the preset term vector database; convert term categories into category vectors and term levels into level vectors; The terminology semantic vector, category vector, and hierarchy vector are concatenated to obtain the option feature vector of each option content; Sort the option feature vectors of each option within the same multi-select field to obtain the option feature vector sequence corresponding to each multi-select field.

5. The large-model-driven integrated resource-demand evaluation system according to claim 4, characterized in that, The methods for constructing the combined feature vector set include: Calculate the vector similarity between any two option feature vectors in the option feature vector sequence corresponding to the same multi-select field, and construct the option similarity matrix; Normalize the vector similarity corresponding to each element in the option similarity matrix to obtain the option correlation degree between each option pair; A correlation weight matrix is ​​constructed based on the correlation between each pair of options. The correlation weight matrix is ​​then fused with the sequence of option feature vectors to obtain the weighted feature vector of each option content. The number of options in each multiple-choice field is counted, and an option number normalization factor is calculated based on the number of options. The weighted feature vectors of each option are summed, and the product of this sum and the option number normalization factor is calculated to obtain the combined feature vector of the corresponding multiple-choice field. The combined feature vectors of all multiple-choice fields are integrated to form a set of combined feature vectors.

6. The large-model-driven integrated resource-demand evaluation system according to claim 5, characterized in that, The methods for constructing the feature alignment matrix include: The classification results include symmetric dimension labels and asymmetric dimension labels. Symmetric dimension labels refer to dimension labels that are the same in both the college profile and the partner unit profile in the profile data. Asymmetric dimension labels refer to dimension labels that exist only in the college profile or the partner unit profile. Obtain the combined feature vector corresponding to each symmetric dimension identifier, and pair the combined feature vector corresponding to the symmetric dimension identifier in the college profile with the combined feature vector corresponding to the collaborating unit profile to obtain the common dimension feature pair; Identify the intermediate dimension identifier corresponding to each asymmetric dimension identifier, and obtain the intermediate semantic vector corresponding to the intermediate dimension identifier. Concatenate the combined feature vector corresponding to the asymmetric dimension identifier with the intermediate semantic vector to obtain the compensation feature vector and add a data source label. Construct a common dimension matrix by arranging all common dimension feature pairs in the order of dimension labels, and construct an asymmetric dimension matrix by arranging each compensation feature vector in the order of dimension labels. The feature alignment matrix is ​​obtained by concatenating the common dimension matrix and the asymmetric dimension matrix; a matching feature space is constructed based on the feature alignment matrix, and the college profile and the partner unit profile are mapped to the matching feature space.

7. The large-model-driven integrated resource-demand evaluation system according to claim 6, characterized in that, The matching and fusion methods include: Identify the collaboration type tags and the effective deadline of the requirements for the collaboration content, and simultaneously call the corresponding dimension weight settings based on the collaboration content type tags; Calculate the time interval between the effective deadline of the demand and the latest collection date, and construct an urgency adjustment coefficient based on the time interval; extract the geographic location weight in the dimension weight setting, and calculate the product of the urgency adjustment coefficient and the geographic location weight to obtain the adjusted geographic location weight; Extract the historical number of collaborations and the most recent collaboration date from the profile data, and calculate the collaboration activity coefficient based on the historical number of collaborations and the most recent collaboration date; In the matching feature space, calculate the dimensional matching value between the cooperation content and each cooperating unit in the dimensions of technology field, technology direction, geographical location and cooperation type. The matching value of each dimension is weighted and fused with the corresponding dimension weight, and the comprehensive matching score of each cooperating unit is obtained by combining the cooperation activity coefficient. The collaborating units are sorted based on their comprehensive matching scores, and those with comprehensive matching scores higher than a preset matching score threshold are selected and arranged into a response processing sequence.

8. The large-model-driven integrated resource-demand evaluation system according to claim 7, characterized in that, The methods for performing location matching and establishing mapping relationships include: Identify the outcome type tags and corresponding outcome field sets of project outcome information, and extract the field names and field contents of the outcome fields; match the field names and field contents with the domain terminology table to determine the standard terms and term categories of the corresponding outcome fields; Construct an evaluation indicator tree based on indicator data, traverse all indicator nodes in the evaluation indicator tree, and read the indicator name, indicator level and related terms of each indicator node. The standard terms of each result field are compared with the indicator-related terms of each indicator node. The indicator nodes that have an intersection between the standard terms and the indicator-related terms are identified as candidate mapping indicator nodes for the corresponding result fields. Based on the indicator hierarchy, candidate mapping indicator nodes for all outcome fields are divided, and the indicator node with the deepest indicator hierarchy is retained as the target mapping indicator node; the mapping relationship between each outcome field and the corresponding target mapping indicator node is established.

9. The large-model-driven integrated resource-demand evaluation system according to claim 8, characterized in that, The methods for performing aggregation calculations include: Based on the mapping relationship and target mapping indicator nodes, all result field sets are grouped to obtain the result field group corresponding to each indicator node; the result field group corresponding to each indicator node is traversed to identify the field type of each result field, including numeric type, option type and text type; Extract the field values ​​of numerical result fields and perform mathematical operations based on preset aggregation rules; extract the field options of option type result fields and convert the field options into corresponding option scores; identify the field content of text type result fields, determine the number of standard terms in the field content based on the domain terminology table, and use the number of standard terms as the text score; The mathematical operation results, option scores, and text scores corresponding to the same indicator node are summed to generate the original indicator score for that indicator node. The original indicator scores of the indicator nodes are normalized to obtain the objective indicator scores. At the same time, the objective indicator scores of all indicator nodes are rewritten into the corresponding indicator nodes in the evaluation indicator tree for updating.

10. The large-model-driven integrated resource-demand evaluation system according to claim 9, characterized in that, The methods for updating the parameters in the performance evaluation process include: Based on the statistical analysis of objective indicator scores and response processing sequence ranking results corresponding to different cooperation contents in the performance feedback record table, the evaluation parameter correction amount is constructed and written into the preset evaluation standard parameter table for updating.

Images