An ai large model-based global college information education consultation method and system

Abstract

This invention provides a method and system for global university information and educational consultation based on an AI-powered large-scale model, relating to the fields of data processing and artificial intelligence. The method includes: collecting official multilingual information from universities worldwide, preprocessing it, and constructing a knowledge graph; fine-tuning a pre-trained large-scale model using multilingual corpora; identifying intent through natural language consultation, and retrieving university lists, admission standards, and related information from the knowledge graph; and providing accurate and personalized educational information consultation services through visualization and formatted recommendations. This invention combines a multilingual knowledge graph, a pre-trained large-scale model, and an AI inference engine, significantly improving the intelligence level of global university information retrieval.

Description

Technical Field

[0001] This invention relates to the fields of data processing and artificial intelligence technology, and in particular to a global school information education consulting method and system based on a large AI model. Background Technology

[0002] With increasing globalization, more and more students and institutions need to search, compare, and analyze information about universities and related programs worldwide to make more accurate educational and career plans. However, in current technology, educational information systems are often scattered across various countries, regions, or institutions, with inconsistent data structures and a lack of support for complex cross-language and cross-disciplinary queries. Specifically: (1) Difficulty in integrating multi-source data Information on educational institutions worldwide is typically distributed across educational databases in different countries or regions, or stored on websites in different languages. Existing systems often struggle to integrate this data uniformly, resulting in limited retrieval efficiency and query accuracy.

[0003] (2) Lack of flexible cross-language support Educational information users often come from all over the world and expect to be able to search and consult in their native language. However, traditional systems often only support one language or a few fixed languages, making it difficult to meet the multilingual needs of users worldwide.

[0004] (3) Insufficient ability to provide personalized in-depth consultation Existing educational information retrieval relies mainly on keyword and simple text searches, lacking the ability to utilize deep learning models to understand user contextual semantics and provide personalized, multi-dimensional analysis results.

[0005] To address the aforementioned issues, traditional solutions often rely on manually or system-configured query statements (such as SQL) to retrieve data, or simplify them to retrieval patterns under fixed templates. This makes it difficult to efficiently provide accurate results when handling multi-table joins or cross-language queries based on user context.

[0006] Therefore, effectively integrating global school data to provide multilingual, highly accurate, and personalized educational information consulting services has become a major technical challenge in the fields of educational information retrieval and artificial intelligence. Summary of the Invention

[0007] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide a global school information education consultation method and system based on AI large model, which combines multilingual knowledge graphs, pre-trained large models and artificial intelligence inference engines to significantly improve the intelligence level of global school information retrieval.

[0008] To achieve the above objectives, the present invention provides the following solution: A global educational information consulting approach for higher education institutions based on a large AI model includes: Collect official information from universities worldwide in multiple languages; The official information is preprocessed and features are extracted to form a knowledge graph; Obtain an open pre-trained large model containing multilingual corpora, and train and fine-tune the open pre-trained large model to obtain a pre-trained language large model; Natural language information is obtained through a front-end page or voice device, and the natural language information is input into the pre-trained language big model for intent recognition to obtain key recognition information; Based on the artificial intelligence reasoning engine, the knowledge graph is invoked to perform retrieval according to the key identification information to obtain retrieval results; the retrieval results include a list of colleges that meet the conditions and their admission standards, as well as the geographical location, language environment, and number of students admitted in previous years associated with the colleges; The search results are then visualized and formatted for recommendation.

[0009] Preferably, the official information includes: the name of the institution, its geographical location, major information, scholarship information, and language requirements; the major information includes the curriculum and length of study.

[0010] Preferably, the official information is preprocessed and features are extracted to form a knowledge graph, including: The official information is deduplicated to obtain the deduplicated text. Based on the context, the deduplicated text is classified to obtain classification data; Based on natural language processing technology, feature extraction is performed on the elements of faculty information, research capabilities, and employment rate in the classification data to construct correlation relationships; A knowledge graph is formed based on the elements and the relationships between them.

[0011] Preferably, based on contextual relationships, the deduplicated text is classified to obtain classification data, including: The deduplicated text is input into the text context relationship feature extraction layer to extract context relationship feature information; The deduplicated text is input into the global feature extraction layer to extract global feature information; The contextual relationship feature information and the global feature information are fused to obtain fused text features; The fused text features are input into the trained text classification model to obtain the classification data.

[0012] Preferably, the deduplicated text is input into the text context relationship feature extraction layer to extract context relationship feature information, including: The initial feature information of the deduplicated text is extracted using a pre-trained language model. The initial feature information is input into the forward gated loop unit and the reverse gated loop unit; The context relationship feature information is obtained by concatenating the outputs of the forward gated loop unit and the reverse gated loop unit.

[0013] Preferably, the outputs of the forward-gated loop unit and the reverse-gated loop unit are concatenated to obtain contextual relationship feature information, including: Formula used: The outputs of the forward-gated loop unit and the reverse-gated loop unit are concatenated to obtain contextual relationship feature information; wherein... This represents the initial feature information input at time t. This represents the output of the positively gated loop unit at time t. This represents the output of the reverse gated loop unit at time t, where GRU stands for gated loop unit and G represents context relation feature information.

[0014] Preferably, the deduplicated text is input into the global feature extraction layer to extract global feature information, including: The initial feature information of the deduplicated text is sequentially input into the convolutional layer and pooling layer to obtain global feature information; wherein, the formula for extracting global feature information is: In the formula, For activation function, For convolution kernel, The kernel size is the convolution kernel size. It is a partial value. This is the feature vector extracted by the i-th convolutional layer. This represents the value of the features extracted by the three different convolutional kernels after passing through the max pooling layer, where C represents the extracted global feature information.

[0015] Preferably, an open pre-trained large model containing multilingual corpora is obtained, and the open pre-trained large model is trained and fine-tuned to obtain a pre-trained language large model, including: Choose an open pre-trained large model containing multilingual corpora; Data deduplication, sentence formatting, and segmentation are performed on a pre-defined multi-corpus sample set to obtain a structured fine-tuned corpus; the multi-corpus sample set includes text information such as study abroad guides, university comparisons, and major interpretations; The open pre-trained large model is supervised using the structured fine-tuning corpus to ensure that the model’s vocabulary and expression capabilities in the field of education are enhanced, thus obtaining a fine-tuned model for basic education scenarios. The basic education scenario fine-tuning model is fine-tuned using pre-set multilingual small sample instruction data to ensure that the model can understand and answer educational information consultation questions expressed in different languages, thus obtaining a pre-trained language model with multilingual reinforcement.

[0016] Preferably, based on an artificial intelligence reasoning engine, a knowledge graph is invoked to perform a retrieval based on the key identification information, and the retrieval results are obtained, including: The key information is mapped to the entity and attribute names of the knowledge graph to construct an executable retrieval request; The search is performed in the knowledge graph according to the executable search request to match the school entities and related attributes that meet the conditions, and to obtain preliminary results; The preliminary results are filtered and sorted based on preset business rules, and information on admission standards, geographical location, and the number of students admitted in previous years is integrated, deduplicated, or merged to obtain the search results; the search results include a list of matching institutions and extended information.

[0017] A global educational information and consulting system for higher education institutions based on an AI-powered large-scale model, comprising: The information collection unit is used to collect official information from universities around the world in multiple languages; The graph construction unit is used to preprocess and extract features from the official information to form a knowledge graph; The model training unit is used to acquire an open pre-trained large model containing multilingual corpora, and to train and fine-tune the open pre-trained large model to obtain a pre-trained language large model. The intent recognition unit is used to obtain natural language information through a front-end page or voice device, and input the natural language information into the pre-trained language big model for intent recognition to obtain key recognition information. The retrieval unit is used to perform retrieval based on the key identification information by calling the knowledge graph based on the artificial intelligence reasoning engine, and to obtain retrieval results. The retrieval results include a list of colleges that meet the conditions and their admission standards, as well as the geographical location, language environment, and number of students admitted in previous years associated with the colleges. The visualization unit is used to visualize and format the search results.

[0018] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects: This invention provides a method and system for global university information education consultation based on an AI-powered large-scale model. The method includes: collecting official information from universities worldwide in multiple languages; preprocessing and extracting features from the official information to form a knowledge graph; acquiring an open pre-trained large-scale model containing multilingual corpora, and training and fine-tuning the open pre-trained large-scale model to obtain a pre-trained language large-scale model; obtaining natural language consultation through a front-end page or voice device, and inputting the natural language consultation into the pre-trained language large-scale model for intent recognition to obtain key recognition information; using an AI inference engine, calling the knowledge graph based on the key recognition information to perform retrieval and obtain search results; the search results include a list of universities that meet the criteria, their admission standards, and related geographical locations, language environments, and past enrollment numbers; and visualizing and formatting recommendations based on the search results. By combining a multilingual knowledge graph, a pre-trained large-scale model, and an AI inference engine, this method provides users with comprehensive, efficient, accurate, and personalized education information consultation services, significantly improving the intelligence level of global university information retrieval and decision-making. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart of the method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the system structure provided in an embodiment of the present invention. Detailed Implementation

[0021] 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.

[0022] The purpose of this invention is to provide a global college information education consultation method and system based on AI large model, which combines multilingual knowledge graph, pre-trained large model and artificial intelligence inference engine to significantly improve the intelligence level of global college information query.

[0023] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1 As shown, this invention provides a global educational consulting method for higher education institutions based on a large AI model, including: Step 100: Collect official information from universities worldwide in multiple languages; Step 200: Preprocess and extract features from official information to form a knowledge graph; Step 300: Obtain an open pre-trained large model containing multilingual corpora, and train and fine-tune the open pre-trained large model to obtain a pre-trained language large model; Step 400: Obtain natural language information through the front-end page or voice device, and input the natural language information into the pre-trained language big data model for intent recognition to obtain key recognition information; Step 500: Based on the artificial intelligence reasoning engine, the knowledge graph is called to perform a search based on the key identification information to obtain the search results; the search results include a list of colleges that meet the conditions and their admission standards, as well as the geographical location, language environment, and number of students admitted in previous years associated with the colleges; Step 600: Visualize and format the search results for recommendations.

[0025] Preferably, the official information includes: the name of the institution, its geographical location, major information, scholarship information, and language requirements; the major information includes the curriculum and length of study.

[0026] Specifically, step 100 in this embodiment includes: Step 101: Identifying Data Sources and Collecting Multilingual Data First, identify the primary sources of global higher education information, including official databases of national education departments, university websites, publicly available data from international education organizations (such as QS rankings and Times rankings), and third-party education consulting platforms. For these sources, develop a multilingual data collection strategy covering mainstream languages ​​(such as English, Chinese, French, and Spanish) to ensure data diversity and coverage. Simultaneously, utilize web scraping technology and API interfaces to automatically collect structured and unstructured data such as university names, geographical locations, program information, scholarship information, and language requirements.

[0027] Step 102: Cleaning and standardization of multilingual data The collected data may contain duplicates, incomplete information, or inconsistent formats, thus requiring data cleaning. Duplicate institution records are eliminated using deduplication algorithms, and multilingual translation tools are used to unify and align institution names and information across different languages. For example, "University of Oxford" and "Oxford University" are identified as the same entity. Furthermore, natural language processing techniques (such as entity recognition and regular expressions) are employed to extract core fields (such as major names, course offerings, and length of study), and the data is stored in a unified structured format (such as JSON or a relational database).

[0028] Step 103: Integration and Storage of Multi-Source Data After cleaning and normalization, the multilingual data is integrated to construct a unified multi-source university database. Entity alignment technology is used to merge descriptions of the same university in different languages ​​and supplement missing information (such as supplementing scholarship or course information from other sources). Finally, the data is stored in the database according to field categories (such as university name, geographical location, and majors offered), providing a high-quality data foundation for the subsequent construction of the knowledge graph.

[0029] Preferably, the official information is preprocessed and features are extracted to form a knowledge graph, including: The official information is deduplicated to obtain the deduplicated text. Based on the context, the deduplicated text is classified to obtain classification data; Based on natural language processing technology, feature extraction is performed on the elements of faculty information, research capabilities, and employment rate in the classification data to construct correlation relationships; A knowledge graph is formed based on the elements and the relationships between them.

[0030] Specifically, step 200 in this embodiment includes: Step 201: Data Deduplication and Normalization First, the collected official information in multiple languages ​​undergoes data deduplication and normalization. String matching algorithms (such as hash algorithms or Jaccard similarity) are used to detect duplicate records, and multilingual translation tools are used to unify the alignment of the same entity described in different languages ​​(e.g., "University of Oxford" and "Oxford University"). For data with inconsistent formats (such as dates, academic years, etc.), normalization is performed using regular expressions or predefined rules to ensure data consistency and accuracy. The deduplicated text data is stored in a structured format (such as JSON or database tables) to lay the foundation for subsequent classification and feature extraction.

[0031] Step 202: Context-based text classification After deduplication, text data is classified using contextual relationships. Pre-trained language models (such as BERT or GPT) are used for semantic analysis, combined with supervised learning methods (such as SVM or classification neural networks) to categorize the text into different types (e.g., faculty information, research capabilities, employment rate). For example, by analyzing keywords (such as "faculty," "research output," "employment rate") and their contextual semantics, the descriptive content is automatically classified into the corresponding topic category. The classified data is stored in a labeled format, providing a clear semantic scope for subsequent feature extraction.

[0032] Step 203: Feature Extraction and Knowledge Graph Construction Based on natural language processing techniques (such as named entity recognition and dependency parsing), key elements (such as the number of professors in faculty information, the number of published papers in research capabilities, and specific percentages in employment rates) are extracted from categorical data. Simultaneously, relation extraction techniques (such as rule-based relation extraction or deep learning models) are used to construct relationships between elements; for example, establishing a triple relationship for "a university - research capabilities - number of published papers". Finally, the extracted elements and relationships are stored in a graph database (such as Neo4j or RDF format) to form a searchable knowledge graph, supporting subsequent inference engine calls.

[0033] Preferably, based on contextual relationships, the deduplicated text is classified to obtain classification data, including: The deduplicated text is input into the text context relationship feature extraction layer to extract context relationship feature information; The deduplicated text is input into the global feature extraction layer to extract global feature information; The contextual relationship feature information and the global feature information are fused to obtain fused text features; The fused text features are input into the trained text classification model to obtain the classification data.

[0034] Preferably, the deduplicated text is input into the text context relationship feature extraction layer to extract context relationship feature information, including: The initial feature information of the deduplicated text is extracted using a pre-trained language model. The initial feature information is input into the forward gated loop unit and the reverse gated loop unit; The context relationship feature information is obtained by concatenating the outputs of the forward gated loop unit and the reverse gated loop unit.

[0035] Preferably, the outputs of the forward-gated loop unit and the reverse-gated loop unit are concatenated to obtain contextual relationship feature information, including: Formula used: The outputs of the forward-gated loop unit and the reverse-gated loop unit are concatenated to obtain contextual relationship feature information; wherein... This represents the initial feature information input at time t. This represents the output of the positively gated loop unit at time t. This represents the output of the reverse gated loop unit at time t, where GRU stands for gated loop unit and G represents context relation feature information.

[0036] Preferably, the deduplicated text is input into the global feature extraction layer to extract global feature information, including: The initial feature information of the deduplicated text is sequentially input into the convolutional layer and pooling layer to obtain global feature information; wherein, the formula for extracting global feature information is: In the formula, For activation function, For convolution kernel, The kernel size is the convolution kernel size. It is a partial value. This is the feature vector extracted by the i-th convolutional layer. This represents the value of the features extracted by the three different convolutional kernels after passing through the max pooling layer, where C represents the extracted global feature information.

[0037] As an example, this embodiment inputs the deduplicated text into a pre-trained language model (such as BERT, GPT, etc.) to extract initial feature information from the text, leveraging its powerful semantic understanding capabilities. These features include the lexical, syntactic, and semantic information of the text, forming the basis for subsequent feature extraction. The initial feature information is then input into forward and backward gated recurrent units (GRUs), respectively. The forward GRU extracts features from the beginning to the end of the text, while the backward GRU extracts features from the end to the beginning. Through bidirectional processing, the dependencies and semantic information between contexts in the text are captured. The outputs of the forward and backward GRUs are concatenated to form complete contextual relationship feature information. This concatenation method comprehensively considers the semantic connections between different parts of the text, providing richer contextual information for classification tasks.

[0038] Furthermore, this embodiment inputs the initial feature information of the text into a convolutional layer, extracting local features (such as patterns in phrases or sentence segments) through convolutional kernels. The diversity of convolutional kernels (such as kernels of different sizes) can capture features of different granularities in the text. The features extracted by the convolutional layer are then input into a pooling layer, where max pooling is used to extract global feature information. The pooling layer effectively reduces feature dimensionality while retaining the most salient feature information, enhancing the model's generalization ability. The pooled feature information represents the global semantic pattern of the text, providing overall semantic support for classification tasks.

[0039] Furthermore, this embodiment fuses contextual relationship features with global features (e.g., through concatenation or weighted fusion) to form fused text features. Contextual features focus on local dependencies within the text, while global features focus on overall semantic patterns; combining the two provides a more comprehensive representation of text features. The fused text features are then input into a trained text classification model (e.g., a fully connected neural network or classifier) ​​to output classification results. Based on the semantic information of the fused features, the classification model categorizes the text into corresponding categories (e.g., faculty information, research capabilities, employment rate, etc.).

[0040] This method innovatively combines contextual features and global features, capturing both local semantic dependencies and extracting overall semantic patterns from text. Contextual features capture semantic relationships between preceding and following sentences using a bidirectional GRU, while global features extract the overall semantic structure of the text through convolution and pooling. The fusion of these two approaches enables the classification model to understand text content more comprehensively. Using a bidirectional GRU to extract contextual features effectively captures semantic dependencies between sentences, making it particularly suitable for complex descriptive texts in the education field (such as descriptions of teacher qualifications or research capabilities). Compared to traditional unidirectional RNNs or LSTMs, bidirectional GRUs are more efficient at capturing long-distance dependencies and contextual information. The introduction of convolutional and pooling layers allows the model to extract global features of different granularities from the text (such as phrase patterns or sentence structures). By combining various convolutional kernels, diverse semantic patterns in the text can be captured, enhancing the robustness and generalization ability of the classification model. The fusion of contextual and global features is the core innovation of this method. Contextual features focus on local dependencies, while global features focus on overall patterns. The combination of the two makes up for the shortcomings of single feature representation, making the classification model more accurate and comprehensive when processing complex text.

[0041] In summary, this method, through the extraction and fusion of contextual and global features, combined with pre-trained language models and deep learning techniques, can efficiently and accurately classify text, making it particularly suitable for complex text processing tasks involving multiple languages ​​and dimensions in the field of education.

[0042] Preferably, an open pre-trained large model containing multilingual corpora is obtained, and the open pre-trained large model is trained and fine-tuned to obtain a pre-trained language large model, including: Choose an open pre-trained large model containing multilingual corpora; Data deduplication, sentence formatting, and segmentation are performed on a pre-defined multi-corpus sample set to obtain a structured fine-tuned corpus; the multi-corpus sample set includes text information such as study abroad guides, university comparisons, and major interpretations; The open pre-trained large model is supervised using the structured fine-tuning corpus to ensure that the model’s vocabulary and expression capabilities in the field of education are enhanced, thus obtaining a fine-tuned model for basic education scenarios. The basic education scenario fine-tuning model is fine-tuned using pre-set multilingual small sample instruction data to ensure that the model can understand and answer educational information consultation questions expressed in different languages, thus obtaining a pre-trained language model with multilingual reinforcement.

[0043] Specifically, step 300 in this embodiment includes: Step 301: Selecting an open pre-trained large model and preprocessing multiple corpus sample sets First, select a large, pre-trained model that supports multiple languages ​​(such as mBERT, XLM-R, or LLaMA). These models have been pre-trained on multilingual corpora and possess cross-lingual semantic understanding capabilities. Next, perform data preprocessing on a multi-corpus sample set in the education field (such as text information like study abroad guides, university comparisons, and major interpretations). This includes deduplication (eliminating duplicate content), sentence formatting (standardizing punctuation, language style, etc.), segmentation (breaking long texts into shorter sentences or paragraphs suitable for model training), and corpus annotation (such as labeling language or topic categories). The result is a structured, fine-tuned corpus, providing a high-quality data foundation for subsequent model fine-tuning.

[0044] Step 302: Supervised fine-tuning to strengthen capacity in the education sector The preprocessed structured fine-tuning corpus was used to supervise the fine-tuning of the open pre-trained large model. By designing education-related tasks (such as classifying school information, generating major interpretations, and answering admission requirement questions), the model was trained to enhance its vocabulary, expression, and semantic understanding capabilities in the education domain. For example, the model improved its understanding of education-specific terms and sentence structures by learning common expressions from the "school comparison" corpus (such as "lower tuition fees" and "higher employment rates"). After fine-tuning, a basic education scenario-based fine-tuned model was obtained, which can initially handle consultation tasks in the education domain.

[0045] Step 303: Fine-tuning small sample instructions to enhance multilingual capabilities Building upon the fine-tuned model for basic education scenarios, further fine-tuning is achieved using pre-set multilingual small-sample instruction data. This small-sample instruction data includes educational information consultation questions and their corresponding answers in different languages ​​(such as English, Chinese, and French), covering common consultation scenarios (such as "recommending suitable universities" or "explaining the employment prospects of a certain major"). Through this small-sample instruction fine-tuning, the model can better understand and answer educational questions expressed in different languages, improving its cross-language generalization ability and answer accuracy. Ultimately, a pre-trained language model with multilingual enhancements is obtained, capable of providing accurate educational information consultation in multilingual environments.

[0046] Optionally, step 400 in this embodiment includes: Step 401: Obtaining input from the front-end page or voice device Users can input natural language information through a front-end page (such as a webpage or mobile application) or a voice device (such as a smart speaker or voice assistant). For text input, the front-end page directly receives the user's text content through a form or chat box; for voice input, the voice device converts speech into text using speech recognition technology (such as Google Speech-to-Text or iFlytek speech recognition). Whether it's text or voice input, it will ultimately generate a piece of natural language text, which will serve as input for subsequent processing.

[0047] Step 402: Preprocessing of Natural Language Text The acquired natural language text needs to be preprocessed to improve the model's understanding. Preprocessing includes removing extra spaces, normalizing punctuation, spell checking (for text input), and language detection (for multilingual input). For example, for a user input "I want to apply to a computer science program in the US, please recommend some schools," the system will normalize it to "I want to apply to a computer science program in the US, please recommend some schools." For multilingual scenarios, preprocessing also needs to identify the input language (such as Chinese, English, etc.) so that the model can select the appropriate language processing path later.

[0048] Step 403: Intent Recognition in Natural Language Consultation The pre-processed natural language text is input into a pre-trained language model for intent recognition. These pre-trained language models (such as GPT and BERT) have been fine-tuned to possess semantic understanding capabilities relevant to the education domain. The model performs semantic parsing on the input text, identifying the user's core intent (such as "recommend universities," "compare majors," "check admission requirements") and related key information (such as target country, major, and language requirements). For example, for the input "I want to apply to a computer science program in the US," the model will identify the intent as "recommend universities," and the key information as "USA" and "computer science major."

[0049] Step 404: Extraction and structuring of key identification information The model's output requires further extraction and structuring to transform key information into a standardized format suitable for subsequent retrieval. For example, "United States" and "Computer Science" are mapped to "Target Country: United States" and "Target Major: Computer Science," respectively. If the user input contains ambiguous information (such as "Recommend some good schools"), the system will complete the missing information using default rules or through user interaction (such as "Which country's schools would you like to apply to?"). Ultimately, the key information is stored in a structured form to support subsequent knowledge graph retrieval.

[0050] Step 405: Multi-round interaction and dynamic adjustment In some cases, a user's natural language inquiry may be incomplete or require further clarification (e.g., "Recommended schools" without specifying a country or major). The system dynamically adjusts its recognition results through multiple rounds of interaction with the user. For example, the system can generate prompting questions (e.g., "Which country's schools do you prefer?") and update key information based on the user's answers. Furthermore, the system can maintain the coherence of the conversation based on context. For instance, when a user asks consecutive questions (e.g., "Recommended schools in the US" followed by "What scholarships are available?"), the system can understand the context and connect the preceding and following questions, ensuring the accuracy and completeness of the recognition results.

[0051] Through the above steps, this embodiment can accurately identify intent and key information from users' natural language inquiries, providing high-quality input for subsequent knowledge graph retrieval and recommendation.

[0052] Preferably, based on an artificial intelligence reasoning engine, a knowledge graph is invoked to perform a retrieval based on the key identification information, and the retrieval results are obtained, including: The key information is mapped to the entity and attribute names of the knowledge graph to construct an executable retrieval request; The search is performed in the knowledge graph according to the executable search request to match the school entities and related attributes that meet the conditions, and to obtain preliminary results; The preliminary results are filtered and sorted based on preset business rules, and information on admission standards, geographical location, and the number of students admitted in previous years is integrated, deduplicated, or merged to obtain the search results; the search results include a list of matching institutions and extended information.

[0053] Specifically, step 500 in this embodiment includes: Step 501: Mapping key information to knowledge graph and constructing retrieval requests First, based on key information extracted from natural language processing (such as target country, major, language requirements, etc.), semantic mapping is performed between this information and entity and attribute names in the knowledge graph. For example, "Computer Science" is mapped to the major entity in the knowledge graph, and "United States" is mapped to the geographical location attribute. Using these mappings, executable search requests (such as SPARQL queries or graph database query instructions) are constructed, specifying the search criteria (such as filtering for target majors or universities in specific regions). This mapping requires the semantic parsing capabilities of an AI inference engine to ensure that key information accurately matches entities and attributes in the knowledge graph.

[0054] Step 502: Perform a search within the knowledge graph to obtain preliminary results. The constructed search request is input into the knowledge graph for execution, matching eligible institution entities and their associated attributes (such as admission standards, geographical location, and past enrollment numbers). The knowledge graph quickly locates relevant information using its triple structure (e.g., "Institution A - Admission Standards - TOEFL 80"), generating preliminary results. These preliminary results may include multiple records, covering eligible institutions and their detailed attributes, providing a foundation for subsequent filtering and integration.

[0055] Step 503: Filtering, sorting, and information integration based on business rules The preliminary results are further processed using pre-defined business rules for filtering, sorting, and information integration. Filtering rules may include excluding institutions that do not meet minimum admission requirements (e.g., those with insufficient language scores) and prioritizing institutions based on user-preferred regions or majors. Sorting rules prioritize institutions based on user-focused metrics (e.g., tuition fees, employment rates, institution rankings). Information integration rules are used to deduplicate or merge duplicate information (e.g., different admission requirements for the same institution). For example, "Institution A - TOEFL 80" and "Institution A - IELTS 6" might be merged into a single complete record. The implementation of these rules ensures the accuracy and usability of the search results.

[0056] As an example, the determination of business rules needs to combine general standards in the education field with user needs. For example, admission criteria, geographical location, and language environment are core elements of common concern, while ranking rules can be dynamically adjusted based on user preferences (e.g., prioritizing institutions with high employment rates or low tuition fees). Furthermore, business rules should have dynamic optimization capabilities, continuously adjusting based on user feedback. For example, if a user repeatedly selects a certain type of institution (such as those known for their research capabilities), this embodiment can automatically adjust the rules to prioritize similar institutions. This dynamic optimization mechanism makes business rules more aligned with actual needs and improves user experience.

[0057] In summary, this embodiment maps key information to a knowledge graph, constructs a search request, executes the search, and integrates it with business rules to ultimately generate a list of matching schools and extended information that meets user needs, thereby achieving precise educational information consulting services.

[0058] As an optional implementation, step 600 of this embodiment includes: Step 601: Structuring and Preparing Search Results After the knowledge graph retrieval is complete, the system returns a list of institutions that meet the criteria, along with their associated attribute information (such as admission standards, geographical location, and the number of students admitted in previous years). This data is typically stored in a structured format (such as JSON or database records). To facilitate visualization and recommendations, the data first needs to be organized and optimized, including deduplication, field completion (such as missing geographical location or language environment information), and data formatting (such as standardizing admission requirements to a score range). The organized data will then be structured according to the user's query intent and preferences (such as sorting by ranking, tuition fees, or employment rate), providing a clear logical structure for subsequent display.

[0059] Step 602: Design and Implementation of Visual Display Search results are displayed graphically on the front-end page to enhance user understanding and interactive experience. Visual design includes the following common forms: (1) Table display: Key information such as school name, geographical location, admission standards, tuition fees, and employment rate are listed in a table format to facilitate quick comparison by users.

[0060] (2) Map display: By marking the geographical location of the institutions on an interactive map, users can intuitively understand the distribution of the target institutions.

[0061] (3) Chart display: Use bar charts, pie charts or line charts to display data such as changes in the number of students admitted to colleges and universities and employment rate trends to help users analyze and make decisions.

[0062] The front-end page can use modern front-end frameworks (such as React or Vue) combined with data visualization libraries (such as ECharts or D3.js) to achieve dynamic and interactive visualization effects.

[0063] Step 603: Generating Formatted Recommendations Building upon the visual presentation, the system generates formatted recommendations based on the user's query intent and preferences. For example, for the query "Recommended US universities for computer science programs," the system will generate a list of recommendations, sorted by ranking or employment rate, with detailed information for each university (e.g., "University Name: Massachusetts Institute of Technology; Location: Boston, USA; Admission Requirements: TOEFL 100 or higher; Employment Rate: 95%"). The recommendations can be displayed in card format, with each card containing core information about the university and relevant images (such as the university logo or campus photos), enhancing the user's reading experience.

[0064] Step 604: Implementation of interactive functions To enhance user experience, the system offers a variety of interactive features, allowing users to filter, sort, and customize recommendation results. For example, users can dynamically adjust recommendation results using filter criteria such as "tuition fees below $30,000" or "employment rate above 90%"; they can also view specific institution information by clicking on markers on the map, or rearrange the institution list by ranking, tuition fees, or other fields using the table header sorting function. Furthermore, the system supports users saving or exporting recommendation results (e.g., generating PDF or Excel files) for later reference and sharing.

[0065] Step 605: Multi-platform adaptation and responsive design To ensure the display of recommendation results adapts to different devices, the system adopts a responsive design, supporting seamless switching between desktop, mobile, and tablet devices. For example, on desktop, users can view tables and maps simultaneously, while on mobile, the system automatically adjusts the layout, separating the tables and maps into switchable views. Furthermore, the system supports multilingual display, allowing users to switch languages ​​(such as Chinese, English, etc.) as needed, ensuring the readability and global applicability of the recommended content. Through multi-platform adaptation and responsive design, the system provides users with a consistent and smooth user experience.

[0066] Through the above steps, this embodiment can display the search results in an intuitive and clear manner, and generate formatted recommended content based on user needs, helping users make efficient decisions and improving the intelligence and personalization of educational information consulting services.

[0067] like Figure 2 As shown, this embodiment, corresponding to the above method, also provides a global educational information consulting system for higher education institutions based on an AI large-scale model, including: The information collection unit is used to collect official information from universities around the world in multiple languages; The graph construction unit is used to preprocess and extract features from the official information to form a knowledge graph; The model training unit is used to acquire an open pre-trained large model containing multilingual corpora, and to train and fine-tune the open pre-trained large model to obtain a pre-trained language large model. The intent recognition unit is used to obtain natural language information through a front-end page or voice device, and input the natural language information into the pre-trained language big model for intent recognition to obtain key recognition information. The retrieval unit is used to perform retrieval based on the key identification information by calling the knowledge graph based on the artificial intelligence reasoning engine, and to obtain retrieval results. The retrieval results include a list of colleges that meet the conditions and their admission standards, as well as the geographical location, language environment, and number of students admitted in previous years associated with the colleges. The visualization unit is used to visualize and format the search results.

[0068] The beneficial effects of this invention are as follows: (1) By constructing a multilingual knowledge graph, this invention integrates information on universities worldwide (including university names, geographical locations, majors, admission requirements, etc.), which solves the problem of scattered and inconsistent information in the prior art and greatly improves query efficiency and information coverage.

[0069] (2) This invention uses a pre-trained large model containing multilingual corpora and enhances its understanding ability in the field of education through fine-tuning, so that the system can support multilingual input and output, meet the educational information consultation needs of global users, and break down language barriers.

[0070] (3) This invention utilizes a pre-trained language big model to perform deep semantic analysis on users’ natural language inquiries, extract key identification information (such as target country, major direction, admission requirements, etc.), and combines it with an artificial intelligence reasoning engine to achieve accurate and efficient knowledge retrieval, providing suggestions on colleges and majors that meet the user’s needs.

[0071] (4) The present invention can dynamically adjust the search results and recommended content according to the user's preferences (such as budget, geographical environment, language requirements, etc.) and context information, provide personalized educational consulting services, and improve the user experience.

[0072] (5) The search results of this invention are presented intuitively in a multimodal format such as tables, comparison charts, and maps, providing users with clear and easy-to-understand information on colleges and universities and admission criteria, which facilitates quick decision-making.

[0073] (6) Through the knowledge graph-based correlation retrieval, the system not only provides a list of colleges that meet the conditions, but also displays in-depth information such as the geographical location, language environment, and number of students admitted in previous years, helping users to evaluate college selection from multiple dimensions.

[0074] (7) This invention continuously optimizes the model and knowledge graph through user feedback and interaction records, gradually improves the ability to understand and meet user needs, and forms a closed loop of sustainable improvement.

[0075] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0076] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A global educational information consulting method for higher education institutions based on an AI large-scale model, characterized in that, include: Collect official information from universities worldwide in multiple languages; The official information is preprocessed and features are extracted to form a knowledge graph; Obtain an open pre-trained large model containing multilingual corpora, and train and fine-tune the open pre-trained large model to obtain a pre-trained language large model; Natural language information is obtained through a front-end page or voice device, and the natural language information is input into the pre-trained language big model for intent recognition to obtain key recognition information; Based on the artificial intelligence reasoning engine, the knowledge graph is invoked to perform retrieval according to the key identification information to obtain retrieval results; the retrieval results include a list of colleges that meet the conditions and their admission standards, as well as the geographical location, language environment, and number of students admitted in previous years associated with the colleges; The search results are then visualized and formatted for recommendation.

2. The global educational consulting method for higher education institutions based on AI large-scale models according to claim 1, characterized in that, The official information includes: institution name, geographical location, major information, scholarship information, and language requirements; the major information includes curriculum and duration of study.

3. The global educational consulting method for higher education institutions based on an AI large-scale model as described in claim 1, characterized in that, The official information is preprocessed and features are extracted to form a knowledge graph, including: The official information is deduplicated to obtain the deduplicated text. Based on the context, the deduplicated text is classified to obtain classification data; Based on natural language processing technology, feature extraction is performed on the elements of faculty information, research capabilities, and employment rate in the classification data to construct correlation relationships; A knowledge graph is formed based on the elements and the relationships between them.

4. The global educational consulting method for institutions based on AI large-scale models according to claim 3, characterized in that, Based on contextual relationships, the deduplicated text is classified to obtain classification data, including: The deduplicated text is input into the text context relationship feature extraction layer to extract context relationship feature information; The deduplicated text is input into the global feature extraction layer to extract global feature information; The contextual relationship feature information and the global feature information are fused to obtain fused text features; The fused text features are input into the trained text classification model to obtain the classification data.

5. The global educational consulting method for higher education institutions based on an AI large-scale model according to claim 4, characterized in that, The deduplicated text is input into the text context relation feature extraction layer to extract context relation feature information, including: The initial feature information of the deduplicated text is extracted using a pre-trained language model. The initial feature information is input into the forward gated loop unit and the reverse gated loop unit; The context relationship feature information is obtained by concatenating the outputs of the forward gated loop unit and the reverse gated loop unit.

6. The global educational consulting method for higher education institutions based on an AI large-scale model as described in claim 5, characterized in that, The outputs of the forward-gated loop unit and the reverse-gated loop unit are concatenated to obtain contextual relationship feature information, including: Formula used: The outputs of the forward-gated loop unit and the reverse-gated loop unit are concatenated to obtain contextual relationship feature information; wherein... This represents the initial feature information input at time t. This represents the output of the positively gated loop unit at time t. This represents the output of the reverse gated loop unit at time t, where GRU stands for gated loop unit and G represents context relation feature information.

7. The global educational consulting method for higher education institutions based on an AI large-scale model as described in claim 6, characterized in that, The deduplicated text is input into the global feature extraction layer to extract global feature information, including: The initial feature information of the deduplicated text is sequentially input into the convolutional layer and pooling layer to obtain global feature information; wherein, the formula for extracting global feature information is: In the formula, For activation function, For convolution kernel, The kernel size is the convolution kernel size. It is a partial value. This is the feature vector extracted by the i-th convolutional layer. This represents the value of the features extracted by the three different convolutional kernels after passing through the max pooling layer, where C represents the extracted global feature information.

8. The global educational consulting method for higher education institutions based on an AI large-scale model according to claim 6, characterized in that, Obtain an open pre-trained large model containing multilingual corpora, and train and fine-tune the open pre-trained large model to obtain a pre-trained language large model, including: Choose an open pre-trained large model containing multilingual corpora; Data deduplication, sentence formatting, and segmentation are performed on a pre-defined multi-corpus sample set to obtain a structured fine-tuned corpus; the multi-corpus sample set includes text information such as study abroad guides, university comparisons, and major interpretations; The open pre-trained large model is supervised using the structured fine-tuning corpus to ensure that the model’s vocabulary and expression capabilities in the field of education are enhanced, thus obtaining a fine-tuned model for basic education scenarios. The basic education scenario fine-tuning model is fine-tuned using pre-set multilingual small sample instruction data to ensure that the model can understand and answer educational information consultation questions expressed in different languages, thus obtaining a pre-trained language model with multilingual reinforcement.

9. The global educational consulting method for higher education institutions based on an AI large-scale model according to claim 6, characterized in that, Based on the artificial intelligence reasoning engine, the knowledge graph is invoked to perform a retrieval based on the key identification information, and the retrieval results are obtained, including: The key information is mapped to the entity and attribute names of the knowledge graph to construct an executable retrieval request; The search is performed in the knowledge graph according to the executable search request to match the school entities and related attributes that meet the conditions, and to obtain preliminary results; The preliminary results are filtered and sorted based on preset business rules, and information on admission standards, geographical location, and number of students admitted in previous years is integrated, deduplicated, or merged to obtain the search results; the search results include a list of matching institutions and extended information.

10. A global educational information consulting system for higher education institutions based on an AI large-scale model, characterized in that: include: The information collection unit is used to collect official information from universities around the world in multiple languages; The graph construction unit is used to preprocess and extract features from the official information to form a knowledge graph; The model training unit is used to acquire an open pre-trained large model containing multilingual corpora, and to train and fine-tune the open pre-trained large model to obtain a pre-trained language large model. The intent recognition unit is used to obtain natural language information through a front-end page or voice device, and input the natural language information into the pre-trained language big model for intent recognition to obtain key recognition information. The retrieval unit is used to perform retrieval based on the key identification information by calling the knowledge graph based on the artificial intelligence reasoning engine, and to obtain retrieval results. The retrieval results include a list of colleges that meet the conditions and their admission standards, as well as the geographical location, language environment, and number of students admitted in previous years associated with the colleges. The visualization unit is used to visualize and format the search results.

Images