College entrance examination voluntary reporting intelligent consultation system based on AI Agent technology

Abstract

The invention relates to the technical field of artificial intelligence and education, and discloses an AI Agent technology-based college entrance examination voluntary reporting intelligent consultation system, which comprises a multi-Agent cooperation module, a domain knowledge graph module, a natural language processing module and a multi-source database module. According to the college entrance examination voluntary reporting intelligent consultation system based on the AI Agent technology, a college entrance examination consultation domain knowledge graph is constructed based on a Neo4j graph database through the domain knowledge graph module, the relationship between entities such as schools and majors and admission batches is stored, multi-dimensional association is efficiently processed through an entity relationship network, university and Major Recovery Agent of the multi-Agent cooperation module is combined with the knowledge graph to screen the schools and the majors, and the college entrance examination voluntary reporting intelligent consultation system based on the AI Agent technology is obtained. The Check Agent verifies data, the Web Search Agent supplements real-time information, and the problems that a traditional database is low in complex relation processing efficiency and cannot recommend in real time are solved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and educational technology, and in particular to an intelligent consulting system for filling out college entrance examination applications based on AI Agent technology. Background Art

[0002] Filling out college entrance examination application forms is a key step in a candidate's academic advancement process, involving multi-dimensional information needs such as university selection, major matching, score line analysis, and employment trend judgment. Candidates and parents need to process massive amounts of dynamic data within a limited time, including enrollment plans, professional training programs, admission data over the years, and changes in the employment market of thousands of universities. Traditional consulting methods are difficult to meet the requirements of efficient and accurate information acquisition, and there is an urgent need for intelligent systems to provide real-time and accurate decision-making support.

[0003] In the existing technology, the college entrance examination volunteer application consultation system mainly faces the following technical bottlenecks: on the one hand, it relies on traditional relational databases to store data such as colleges, majors, and admission rules. Due to the lack of structured modeling capabilities for complex relationships between entities (such as the multi-dimensional association between school-major-admission batch-employment direction), the data association efficiency is low during multi-condition queries and cannot meet real-time recommendation needs; on the other hand, the natural language processing model based on keyword matching has difficulty in parsing the deep semantics of questions. For example, it is unable to recognize the implicit logical relationship between scores, elective subjects, and major categories in the question "Which computer majors can I apply for with 580 points in physics, chemistry, and biology?", resulting in a deviation between the consultation results and the user's actual needs. In addition, the single module architecture makes it difficult to dynamically integrate external information such as new enrollment policies and employment data, and the system scalability is limited. Therefore, an intelligent college entrance examination volunteer application consultation system based on AI Agent technology is proposed. Summary of the Invention

[0004] (1) Technical problems solved

[0005] In response to the shortcomings of the existing technology, the present invention provides an intelligent consulting system for filling out college entrance examination applications based on AI Agent technology. Through multi-agent collaboration, domain knowledge graphs and deep learning models, it improves consulting efficiency and accuracy, and solves the technical problems of the existing college entrance examination application consulting system, such as the difficulty of traditional relational databases in efficiently processing complex relationships between entities such as colleges and majors, the inability of natural language processing models based on keyword matching to accurately parse the implicit requirements of questions, and the insufficient scalability of the system.

[0006] (2) Technical solution

[0007] To achieve the above-mentioned goal of improving consultation efficiency and accuracy through multi-agent collaboration, domain knowledge graphs, and deep learning models, the present invention provides the following technical solution: an intelligent consultation system for filling out college entrance examination applications based on AI Agent technology, comprising:

[0008] Multi-agent collaboration module: Contains seven AI agents with different functions, namely Plan Agent, School Agent, Major Agent, Web Search Agent, University and Major Recommendation Agent, Check Agent, and Answer Agent;

[0009] Domain knowledge graph module: A knowledge graph for the college entrance examination consulting field built based on the Neo4j graph database, storing entities and relationships. The entities include schools, majors, institutions, time, location, and problem focus; the relationships include admission batches, score difference, enrollment plan, and affiliated institutions;

[0010] Natural language processing module: includes question classification model and entity recognition model, including:

[0011] The question classification model is a C&W-CNN-BiLSTM-Attention recurrent neural network model, which uses a dual-channel structure of character vectors and word vectors. The character vector branch and the word vector branch extract phrase-level features through CNN respectively, and combine the BiLSTM and Attention mechanisms to fuse global features and key features.

[0012] The entity recognition model is a BERT-BiLSTM-CRF model, which generates word vectors containing contextual semantics based on the BERT pre-trained model, extracts sequence features through bidirectional LSTM, and combines the CRF layer to achieve accurate recognition of entity boundaries and categories;

[0013] Multi-source database module: including school information database, professional information database, admissions database and recruitment database. The database obtains structured and unstructured data from university official websites and education platforms through crawler technology, and stores them after manual labeling and structured processing.

[0014] Preferably, the Plan Agent is used to perform the following steps:

[0015] Receive user natural language questions and analyze the question intent through the natural language processing module;

[0016] Based on the entity relationship network of the knowledge graph, a multi-stage task sequence is generated, and the task sequence includes ranking query, school screening, professional information extraction, and employment data acquisition.

[0017] Preferably, the University and Major Recommendation Agent is used to:

[0018] Based on the user's score and subject combination, query the provincial ranking through the admissions database;

[0019] Based on the ranking and the relationship between admission batches and score differences in the knowledge graph, select the schools and majors that can be applied for;

[0020] Combine the school strength data returned by SchoolAgent and the major details returned by MajorAgent to generate a candidate recommendation list.

[0021] Preferably, the specific structure of the C&W-CNN-BiLSTM-Attention model includes:

[0022] Word vector layer: The CBOW model is used to generate word vectors with a dimension of 200. The input is the question sequence after word segmentation.

[0023] Word vector layer: Use LTP word segmentation tool to perform word segmentation, and generate word vectors through CBOW model with a dimension of 200;

[0024] Convolutional layer: Apply a convolution kernel of size 3 to the character vector matrix and word vector matrix respectively to extract phrase-level features;

[0025] Bidirectional LSTM layer: performs time series modeling on the feature matrix output by the convolutional layer, with 100 hidden units;

[0026] Attention layer: performs weighted summation on the feature vectors output by LSTM to generate feature representations containing key information;

[0027] Fusion layer: The outputs of the character vector branch and the word vector branch are fused through the attention mechanism and input into the Softmax layer to complete the classification.

[0028] Preferably, the specific process of the BERT-BiLSTM-CRF model includes:

[0029] Input layer: Split the question into words, add [CLS] and [SEP] labels, and generate 768-dimensional word vectors using the BERT-base pre-trained model, including positional encoding and contextual semantics.

[0030] Bidirectional LSTM layer: performs forward and backward encoding on the word vector sequence and outputs a feature vector containing the context information;

[0031] CRF layer: Based on the feature vector output by BiLSTM and the transfer matrix, the optimal label sequence is calculated. The labels are labeled using the BIO method, and the categories include school, major, time, and location.

[0032] Preferably, the Check Agent is used to:

[0033] Perform consistency checks on the data returned by each agent, including the matching of school admission data with enrollment plans, and the relevance of professional curriculum settings with training objectives;

[0034] Through the reasoning mechanism of the knowledge graph, the logical rationality of the recommendation results is verified and abnormal data of "admission score line lower than batch line" is filtered out.

[0035] Preferably, the Web Search Agent is used to:

[0036] When the knowledge graph and database lack the required data, the latest enrollment policies and employment market trends can be obtained in real time through web crawlers;

[0037] Structured data extraction technology is used to extract key information from web pages, which is then cleaned and stored in a temporary knowledge base for other agents to access.

[0038] Preferably, the Answer Agent is used to:

[0039] Generate structured answers based on multi-source data verified by Check Agent, including school recommendation lists, major details cards, score trend charts, and job information;

[0040] Template generation technology is used to map data into a natural language answer framework to ensure the readability and information integrity of the answers.

[0041] Preferably, the multi-agent collaboration process is as follows:

[0042] The user enters a question - the Plan Agent parses and decomposes the task - the University and MajorRecommendation Agent is called to generate a candidate list - the School Agent and Major Agent add details - the JobAgent obtains employment data - the Web Search Agent adds missing information - the Check Agent verifies the data - the Answer Agent generates an answer.

[0043] Preferably, the method for constructing the knowledge graph includes:

[0044] Model layer construction: define entity types, attributes, and relationships, using a top-down design approach. Entity types include schools and majors; attributes include admission scores and course settings; and relationships include belonging to and offering.

[0045] Data layer filling: Use Python to call the py2neo library to connect to the Neo4j database, import structured data in the form of triples, and form a network structure of nodes and edges;

[0046] Visual display: Display entity relationship diagrams through the Neo4j front-end interface, supporting node expansion, relationship query and attribute filtering.

[0047] (3) Beneficial effects

[0048] Compared with the existing technology, the present invention provides an intelligent consulting system for filling out college entrance examination applications based on AI Agent technology, which has the following beneficial effects:

[0049] 1. This AI Agent-based intelligent consulting system for college entrance examination applications builds a knowledge graph for college entrance examination consultation based on the Neo4j graph database through the domain knowledge graph module. It stores entities such as schools and majors, as well as relationships such as admission batches, and uses entity relationship networks to efficiently process multi-dimensional associations. The University and Major Recommendation Agent in the multi-agent collaboration module combines the knowledge graph to screen schools and majors, the Check Agent verifies data, and the Web Search Agent supplements real-time information. This solves the problem of traditional databases being inefficient in processing complex relationships and unable to make real-time recommendations.

[0050] 2. This AI Agent-based intelligent consulting system for college entrance examination applications uses the C&W-CNN-BiLSTM-Attention model in the natural language processing module to extract word vectors using dual channels, fuses features using the BiLSTM and Attention mechanisms, and uses the BERT-BiLSTM-CRF model to generate word vectors containing contextual semantics for accurate entity recognition. Furthermore, the multi-agent collaboration module uses independent services and message queue communication for each agent, while the Web Search Agent dynamically acquires external data. This addresses the issues of inaccurate semantic parsing in natural language processing and insufficient system scalability. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is the architecture diagram of the intelligent consulting system for filling out college entrance examination applications of the present invention. DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the embodiments of the present invention and the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0053] See also Figure 1 , an intelligent consulting system for filling out college entrance examination applications based on AI Agent technology, including:

[0054] Multi-agent collaboration module: Contains seven AI agents with different functions, namely Plan Agent, School Agent, Major Agent, Web Search Agent, University and Major Recommendation Agent, Check Agent, and Answer Agent;

[0055] Domain knowledge graph module: A knowledge graph for the college entrance examination consulting field built based on the Neo4j graph database, storing entities and relationships. The entities include schools, majors, institutions, time, location, and problem focus; the relationships include admission batches, score difference, enrollment plan, and affiliated institutions;

[0056] Natural language processing module: includes question classification model and entity recognition model, including:

[0057] The question classification model is a C&W-CNN-BiLSTM-Attention recurrent neural network model, which uses a dual-channel structure of character vectors and word vectors. The character vector branch and the word vector branch extract phrase-level features through CNN respectively, and combine the BiLSTM and Attention mechanisms to fuse global features and key features.

[0058] The entity recognition model is a BERT-BiLSTM-CRF model, which generates word vectors containing contextual semantics based on the BERT pre-trained model, extracts sequence features through bidirectional LSTM, and combines the CRF layer to achieve accurate recognition of entity boundaries and categories;

[0059] Multi-source database module: including school information database, professional information database, admissions database and recruitment database. The database obtains structured and unstructured data from university official websites and education platforms through crawler technology, and stores them after manual labeling and structured processing.

[0060] Example 1:

[0061] This embodiment focuses on the overall system architecture and starts from the technical implementation of multi-agent collaboration mechanism, knowledge graph construction and natural language processing model.

[0062] The multi-agent collaboration module adopts a distributed microservice architecture and achieves elastic scaling through container orchestration technology. Each agent is encapsulated as an independent service unit and equipped with dedicated computing resources and memory space. Asynchronous communication is achieved through message queues. The message format follows the JSON Schema specification and includes fields such as task type, parameter set, priority, and callback address. The task scheduling core determines the execution order based on a directed acyclic graph algorithm and uses an in-memory database to cache intermediate results to achieve millisecond-level data response.

[0063] The construction of the domain knowledge graph is divided into the model layer and the data layer. The model layer uses the ontology engineering method to define 7 core entities and 12 relationship types. The data layer extracts entities and relationships from multi-source heterogeneous data through ETL tools, and adopts an incremental import strategy to build a basic graph skeleton. The entity similarity is calculated through the graph embedding algorithm to identify duplicate entities and merge them. Finally, the attribute graph model is used to store the complete knowledge network and establish a composite index to improve the performance of complex queries.

[0064] The natural language processing module adopts a multimodal fusion architecture. The word vector branch of the question classification model generates context-related representations through a pre-trained language model. The word vector branch combines domain dictionaries and syntactic analysis techniques to capture professional terminology and semantic associations. The two types of vectors extract local features through convolutional neural networks, and then model sequence dependencies through a bidirectional long short-term memory network. The attention mechanism dynamically assigns weights to highlight key information fragments, and finally achieves intent classification through a fully connected layer.

[0065] Example 2:

[0066] This embodiment focuses on the question parsing and task generation mechanism of Plan Agent, and details the process of intent recognition and task sequence construction.

[0067] After receiving user questions, Plan Agent performs a text preprocessing pipeline operation, removes redundant symbols through a regular expression engine, uses a language model to convert between simplified and traditional Chinese characters and perform word segmentation, and combines a domain dictionary to identify educational terms. Intent parsing uses an ensemble learning strategy, integrating the results of a rule-based classifier and a deep learning classifier, and determines the final intent category through a voting mechanism.

[0068] Task sequence generation is based on semantic reasoning of the knowledge graph. Taking the "school recommendation" intent as an example, the initial task node is first determined by querying the "school-major-admission batch" association path through the graph database. The database stored procedure is called based on the user context information to calculate the provincial ranking. Candidate batches are generated based on the ranking range and admission rules. The semantic matching algorithm is used to analyze the correlation between the intent and the knowledge graph entity, and the task chain is dynamically expanded to form a complete task dependency graph.

[0069] Example 3:

[0070] This embodiment focuses on the recommendation logic of the recommendation agent and describes the technical details of ranking matching and candidate list generation.

[0071] After receiving the user's scores and subject selection information, the recommendation agent accesses the admissions database, performs distributed query operations, locates the user's scores in the provincial score database, calculates the relative ranking through window functions, constructs the ranking fluctuation range based on the historical three-year admission data, and screens the set of colleges and universities that meet the ranking requirements based on the admission rules in the knowledge graph, applying multi-dimensional filtering strategies such as regional preference, subject strength, and professional matching.

[0072] The candidate list is generated using a hybrid sorting algorithm. The basic sorting is based on the comprehensive ranking of colleges and professional ratings. The personalized sorting is combined with the user's historical interaction data. The real-time sorting takes into account the dynamic changes in the enrollment plan. The similarity calculation model is used to evaluate the match between each candidate college and user needs. Finally, a recommendation list is generated that includes information such as school code, professional name, admission score lines in the past three years, and ranking fluctuation trends.

[0073] Example 4:

[0074] This embodiment describes the consistency verification and logical reasoning process in detail for the Check Agent data verification mechanism.

[0075] Data verification adopts a multi-dimensional verification strategy. Integrity verification checks whether required fields are missing, compares data structures through predefined metadata models, and consistency verification is achieved through the constraint rules of the knowledge graph. It checks the logical relationship between the school's enrollment plan and the actual number of admissions, as well as whether the professional course settings cover the core knowledge areas required by the training objectives.

[0076] Logical reasoning is implemented based on descriptive logic, building a semantic rule library in the field of education. It defines rules such as "the admission score must not be lower than the batch score" and "the professional subject selection requirements must include physics or chemistry". It performs forward chain reasoning through the inference engine, automatically detects data conflicts and logical contradictions, generates verification reports and marks abnormal data to ensure the reliability of recommendation results.

[0077] Embodiment 5:

[0078] This embodiment focuses on the real-time data acquisition capability of the Web Search Agent and describes in detail the web crawler and data structuring technology.

[0079] The web crawler system adopts a distributed architecture, deploys multiple crawler nodes to form a cluster, designs special crawler strategies for different data sources, uses depth-first traversal for university official websites, incremental crawling for education platforms, and scheduled full updates for policy document websites. It implements IP rotation and request header camouflage through proxy pools and User-Agent pools, and configures request frequency control strategies to avoid triggering the anti-crawling mechanism of the target website.

[0080] Data structuring processing is divided into three stages. First, entities and relationships are identified through natural language processing technology, and entities are extracted from the admissions policy text. Then, the template matching algorithm is used to convert the unstructured text into a key-value pair format. Finally, the knowledge graph alignment technology is used to integrate the newly extracted data with the existing knowledge network to update the temporary knowledge base.

[0081] Example 6:

[0082] This embodiment takes a specific question as an example to describe in detail the technical execution process and data interaction details of multi-agent collaboration.

[0083] When a user enters "I scored 640 in physics, chemistry, and biology and want to study computer science, which schools do you recommend?", PlanAgent interprets the question as "school and major recommendations" and generates an execution sequence consisting of six task nodes. After receiving the task, the recommendation agent accesses the provincial admissions database, quickly locates the user's score through a distributed index, and calculates their provincial ranking as 8,500. Combining the admissions rules in the knowledge graph, it selects a set of colleges and universities in the first batch of undergraduate programs that offer computer science majors and require physics, chemistry, and biology as elective subjects.

[0084] SchoolAgent uses the knowledge graph to query the comprehensive strength indicators of candidate institutions, including the results of the Ministry of Education's subject evaluation, scientific research funding, and the number of national laboratories. Major Agent extracts computer science curriculum, training programs, and graduate destination data. Web Search Agent obtains the latest enrollment dynamics and policy adjustment information of target institutions in real time. Check Agent verifies the consistency and logic of all data and confirms that the admission ranking fluctuations of computer science majors in the recommended institutions over the past three years are within a reasonable range.

[0085] Answer Agent integrates all data and calls the template engine to generate structured answers. It first displays a list of recommended schools, including their comprehensive rankings, professional ratings, and the admission scores of the past three years. It then provides a detailed introduction to the core courses and career prospects of computer science majors. Finally, it embeds a data visualization component to display the target school's enrollment plan changes and admission ranking fluctuation curve in the user's province over the years. The entire processing flow is implemented through an asynchronous messaging mechanism, with each agent processing tasks in parallel, ultimately completing all operations and returning the results within 1.2 seconds.

[0086] In summary, this intelligent consulting system for filling out college entrance examination applications based on AI Agent technology builds a knowledge graph for college entrance examination consultation based on the Neo4j graph database through the domain knowledge graph module, stores entities such as schools and majors and relationships such as admission batches, and uses the entity relationship network to efficiently process multi-dimensional associations. The University and Major Recommendation Agent of the multi-agent collaboration module combines the knowledge graph to screen schools and majors, the Check Agent verifies data, and the Web Search Agent supplements real-time information, solving the problem that traditional databases are inefficient in processing complex relationships and cannot make real-time recommendations.

[0087] In addition, this intelligent consulting system for filling out college entrance examination applications based on AI Agent technology uses the C&W-CNN-BiLSTM-Attention model word vector dual-channel feature extraction of the natural language processing module, the BiLSTM and Attention mechanism fusion feature, and the BERT-BiLSTM-CRF model to generate word vectors with contextual semantics to achieve accurate entity recognition. In addition, each agent in the multi-agent collaboration module provides independent services and message queue communication, and the Web Search Agent dynamically obtains external data. This solves the problems of inaccurate semantic parsing of natural language processing and insufficient system scalability. It also solves the technical problems of the existing college entrance examination application consulting system, such as the difficulty of traditional relational databases in efficiently processing complex relationships between entities such as colleges and majors, the inability of natural language processing models based on keyword matching to accurately parse the implicit requirements of questions, and the insufficient system scalability.

[0088] The relevant modules involved in this system are all hardware system modules or functional modules that combine computer software programs or protocols with hardware in the existing technology. The computer software programs or protocols involved in the functional modules are themselves technologies that are well known to those skilled in the art and are not improvements to this system. The improvements to this system are the interaction or connection relationships between the modules, that is, improvements to the overall structure of the system to solve the corresponding technical problems to be solved by this system.

[0089] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. The intelligent consulting system for filling out college entrance examination applications based on AI Agent technology is characterized by: include: Multi-agent collaboration module: Contains seven AI agents with different functions, namely Plan Agent, School Agent, Major Agent, Web Search Agent, University and Major Recommendation Agent, CheckAgent, and Answer Agent; Domain knowledge graph module: A knowledge graph for the college entrance examination consulting field built based on the Neo4j graph database, storing entities and relationships. The entities include schools, majors, institutions, time, location, and problem focus; the relationships include admission batches, score difference, enrollment plan, and affiliated institutions; Natural language processing module: includes question classification model and entity recognition model, including: The question classification model is a C&W-CNN-BiLSTM-Attention recurrent neural network model, which uses a dual-channel structure of character vectors and word vectors. The character vector branch and the word vector branch extract phrase-level features through CNN respectively, and combine the BiLSTM and Attention mechanisms to fuse global features and key features. The entity recognition model is a BERT-BiLSTM-CRF model, which generates word vectors containing contextual semantics based on the BERT pre-trained model, extracts sequence features through bidirectional LSTM, and combines the CRF layer to achieve accurate recognition of entity boundaries and categories; Multi-source database module: including school information database, professional information database, admissions database and recruitment database. The database obtains structured and unstructured data from university official websites and education platforms through crawler technology, and stores them after manual labeling and structured processing.

2. The intelligent consulting system for filling out college entrance examination applications based on AI Agent technology according to claim 1 is characterized in that: The Plan Agent is used to perform the following steps: Receive user natural language questions and analyze the question intent through the natural language processing module; Based on the entity relationship network of the knowledge graph, a multi-stage task sequence is generated, and the task sequence includes ranking query, school screening, professional information extraction, and employment data acquisition.

3. The intelligent consulting system for filling out college entrance examination applications based on AI Agent technology according to claim 1 is characterized in that: The University and Major Recommendation Agent is used to: Based on the user's score and subject combination, query the provincial ranking through the admissions database; Based on the ranking and the relationship between admission batches and score differences in the knowledge graph, select the schools and majors that can be applied for; Combine the school strength data returned by the School Agent and the major details returned by the Major Agent to generate a candidate recommendation list.

4. The intelligent consulting system for filling out college entrance examination applications based on AI Agent technology according to claim 1 is characterized in that: The specific structure of the C&W-CNN-BiLSTM-Attention model includes: Word vector layer: The CBOW model is used to generate word vectors with a dimension of 200. The input is the question sequence after word segmentation. Word vector layer: Use LTP word segmentation tool to perform word segmentation, and generate word vectors through CBOW model with a dimension of 200; Convolutional layer: Apply a convolution kernel of size 3 to the character vector matrix and word vector matrix respectively to extract phrase-level features; Bidirectional LSTM layer: performs time series modeling on the feature matrix output by the convolutional layer, with 100 hidden units; Attention layer: performs weighted summation on the feature vectors output by LSTM to generate feature representations containing key information; Fusion layer: The outputs of the character vector branch and the word vector branch are fused through the attention mechanism and input into the Softmax layer to complete the classification.

5. The intelligent consulting system for filling out college entrance examination applications based on AI Agent technology according to claim 1 is characterized in that: The specific process of the BERT-BiLSTM-CRF model includes: Input layer: Split the question into words, add [CLS] and [SEP] labels, and generate 768-dimensional word vectors using the BERT-base pre-trained model, including positional encoding and contextual semantics. Bidirectional LSTM layer: performs forward and backward encoding on the word vector sequence and outputs a feature vector containing the context information; CRF layer: Based on the feature vector output by BiLSTM and the transfer matrix, the optimal label sequence is calculated. The labels are labeled using the BIO method, and the categories include school, major, time, and location.

6. The intelligent consulting system for filling out college entrance examination applications based on AI Agent technology according to claim 1 is characterized in that: The Check Agent is used to: Perform consistency checks on the data returned by each agent, including the matching of school admission data with enrollment plans, and the relevance of professional curriculum settings with training objectives; Through the reasoning mechanism of the knowledge graph, the logical rationality of the recommendation results is verified, and abnormal data such as "admission score line is lower than the batch line" is filtered out.

7. The intelligent consulting system for filling out college entrance examination applications based on AI Agent technology according to claim 1 is characterized in that: The Web Search Agent is used to: When the knowledge graph and database lack the required data, the latest enrollment policies and employment market trends can be obtained in real time through web crawlers; Structured data extraction technology is used to extract key information from web pages, which is then cleaned and stored in a temporary knowledge base for other agents to access.

8. The intelligent consulting system for filling out college entrance examination applications based on AI Agent technology according to claim 1 is characterized in that: The Answer Agent is used to: Generate structured answers based on multi-source data verified by Check Agent, including school recommendation lists, major details cards, score trend charts, and job information; Template generation technology is used to map data into a natural language answer framework to ensure the readability and information integrity of the answers.

9. The intelligent consulting system for filling out college entrance examination applications based on AI Agent technology according to claim 1 is characterized in that: The multi-agent collaboration process is as follows: The user enters a question - the Plan Agent parses and decomposes the task - the University and MajorRecommendation Agent is called to generate a candidate list - the School Agent and Major Agent add details - the JobAgent obtains employment data - the Web Search Agent adds missing information - the Check Agent verifies the data - the Answer Agent generates an answer.

10. The intelligent consulting system for filling out college entrance examination applications based on AI Agent technology according to claim 1 is characterized in that: The method for constructing the knowledge graph includes: Model layer construction: define entity types, attributes, and relationships, using a top-down design approach. Entity types include schools and majors; attributes include admission scores and course settings; and relationships include belonging to and offering. Data layer filling: Use Python to call the py2neo library to connect to the Neo4j database, import structured data in the form of triples, and form a network structure of nodes and edges; Visual display: Display entity relationship diagrams through the Neo4j front-end interface, supporting node expansion, relationship query and attribute filtering.

Images