Academic group mining and directional recommendation system for regional natural fund

Abstract

The invention discloses an academic group mining and directional recommendation system for regional natural fund, and relates to the technical field of computers. The academic group mining and directional recommendation system for the regional natural fund comprises a data management system, a knowledge graph module, a mixed recommendation algorithm module, an AI robot module and a user interaction interface module, and a multi-dimensional heterogeneous relation graph, an innovative cold start solution and a lightweight graph recommendation framework are constructed. Accurate mining, intelligent recommendation and efficient management of scientific researchers and projects are realized. According to the method, potential interests and cross-domain association requirements of users can be deeply captured, the problem of data sparsity in a cold start stage is effectively solved, and rapid deployment and real-time recommendation are realized through lightweight design. And powerful technical support and decision basis are provided for reasonable distribution and scientific research and innovation of regional natural funds.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an academic group mining and directional recommendation system for regional natural science funds. Background Art

[0002] In today's scientific research landscape, regional natural science foundations play a crucial role in promoting the vigorous development of regional scientific research and provide a crucial financial guarantee for the smooth implementation of numerous research projects within the region. The rational and precise allocation and deployment of regional natural science foundation resources has profound and critical strategic significance for improving research efficiency, stimulating academic innovation, and promoting the overall advancement of regional scientific research.

[0003] However, with the vigorous development of scientific research, the number of researchers has continued to rise, the number and scale of research projects have also expanded, and project information has become increasingly complex and diverse. At the same time, various academic communities have sprung up like mushrooms after rain, enriching the research ecosystem. Against this backdrop, traditional approaches to managing researchers and projects have exposed numerous shortcomings that hinder their adaptability to the new landscape. Traditional management models often rely on relatively rigid, single processes and standards, making it difficult to accurately capture the unique needs of each researcher, the individual characteristics of each research project, and the diverse demands of different academic communities. During the project application process, a lack of effective information integration and analysis mechanisms may cause some high-quality projects to miss out on funding opportunities. During the project implementation phase, inflexible tracking of researchers and resource allocation prevents timely adjustments to project progress, resulting in wasted resources or hindered progress. During project completion and outcome evaluation, it is difficult to comprehensively and accurately measure the actual value and impact of project achievements, hindering the optimization and improvement of subsequent research work.

[0004] Traditional recommendation systems often rely on a single user-item interaction graph, capturing only direct behavioral associations (e.g., clicks and views). They struggle to integrate heterogeneous information across multiple dimensions, such as social relationships, academic fields, and professional hierarchies. This results in superficial recommendations, lacking a deep understanding of users' potential interests. In cold-start scenarios, new users lack historical interaction data, and traditional collaborative filtering algorithms are limited by data sparsity, resulting in low-quality recommendations. Furthermore, content recommendations rely on user feedback, lacking in diversity and failing to meet new users' need for cross-domain information exploration.

[0005] Therefore, it is urgent to establish a new model for scientific research personnel and project management that is more scientific, efficient, and accurate, so as to fully adapt to the current complex and changing scientific research environment, achieve the optimal allocation of regional natural fund resources, and provide solid and powerful support for the continued prosperity of regional scientific research. In response to the shortcomings of the existing technology, the present invention provides an academic group mining and targeted recommendation system for regional natural fund to solve the above problems. Summary of the Invention

[0006] In response to the shortcomings of existing technologies, the present invention provides an academic community mining and targeted recommendation system for regional natural science funds. By constructing a multi-dimensional heterogeneous relationship graph, an innovative cold start solution, and a lightweight graph recommendation framework, it achieves precise mining, intelligent recommendation, and efficient management of scientific researchers and projects. It can deeply capture users' potential interests and cross-domain related needs, effectively solve the data sparsity problem in the cold start phase, and achieve rapid deployment and real-time recommendations through a lightweight design. It provides strong technical support and decision-making basis for the rational allocation of regional natural science funds and scientific research innovation.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: a regional natural fund academic group mining and targeted recommendation system, comprising:

[0008] Data management system, including:

[0009] A database storage unit is used to store basic information of scientific researchers, project information and team collaboration relationship data using a relational database;

[0010] Multi-field joint search unit, used to perform cross-field joint query in the researcher information table and project information table through keyword fuzzy search, and return the search results in tabular form;

[0011] Knowledge graph module, including:

[0012] Neo4j graph database storage unit, used to store graph structure data of scientific researcher nodes, project nodes, their collaboration relationships, and professional affiliation relationships;

[0013] Echarts visualization engine, used to dynamically generate researcher-project relationship maps, supporting node dragging and hierarchical zooming operations;

[0014] Hybrid recommendation algorithm module, including:

[0015] The three-layer relationship graph construction unit is used to generate user-user team relationship edges based on the team collaboration table, generate user-user research relationship edges based on research directions, and generate user-professional category edges through professional classification mapping files;

[0016] The cold start processing unit is used to search for projects in the same field according to their professional categories when the user has no historical interaction data, and integrate the real-time search keyword matching results;

[0017] AI robot module, including:

[0018] Spark 4.0 Ultra large model interaction unit, used to parse user natural language requests and generate text responses;

[0019] Keyword search unit, used to segment user input text and perform multi-field matching search in Chinese keywords, project names, and subject fields in the project database;

[0020] The user interaction interface module includes an independent functional partition navigation bar, which is respectively associated with the interactive entrances of data management, knowledge graph display, recommendation result list and AI dialogue window.

[0021] Preferably, the database storage unit adopts a MySQL relational database, and the storage unit includes a scientific researcher table, a project information table and a team leader-member relationship table.

[0022] Preferably, the multi-field joint retrieval unit supports performing OR logical joint queries on the name, gender, degree, title, unit name, and major fields of the scientific researcher table.

[0023] Preferably, the knowledge graph module includes:

[0024] An automated graph construction unit that uses the pandas library to parse CSV-formatted researcher collaboration data files;

[0025] The node relationship batch creation unit uses the py2neo library to batch import the parsed data into the Neo4j database.

[0026] Preferably, the node relationship batch creation unit includes:

[0027] The personnel collaboration relationship creates a sub-unit, mapping the relationship between the team leader and the members to the "team leader-belongs" relationship edge;

[0028] The scientific research participation relationship creates a sub-unit, mapping the association between personnel and projects as a "participation" relationship edge.

[0029] Preferably, the three-layer relationship graph construction unit includes:

[0030] The team collaboration subgraph generation module constructs a relationship edge with type='team' based on the team leader ID and member ID fields;

[0031] The professional hierarchical mapping module associates the user's research direction with the predefined professional category nodes through the professional classification mapping file.

[0032] Preferably, the cold start processing unit includes:

[0033] The professional field search submodule executes SQL queries to obtain projects that users in the same professional category are responsible for;

[0034] The behavioral data fusion submodule calls the TF-IDF vectorizer to calculate the cosine similarity between user search keywords and project summaries.

[0035] Preferably, the AI ​​robot module includes:

[0036] Context memory storage, used to save the context state of the user's continuous conversation;

[0037] The multi-field search executor performs keyword matching in the priority order of Chinese keywords → project name → first-level discipline → third-level discipline → project summary.

[0038] Preferably, the user interaction interface module is constructed using the Django framework, and the interface module includes:

[0039] The scientific research personnel management function area provides an interface for adding, deleting, modifying and checking personnel information;

[0040] Knowledge graph display area, dynamically loading the scientific research relationship network diagram generated by Echarts;

[0041] The recommendation result display area displays a list of recommended items based on graph algorithms and content matching in columns.

[0042] Preferably, the system deployment architecture includes:

[0043] Django application server, used to process HTTP requests and call various functional modules;

[0044] Parallel MySQL database and Neo4j graph database, storing structured data and graph relational data respectively;

[0045] NetworkX graph computing engine, used to process user similarity calculation requests in real time.

[0046] The present invention discloses an academic group mining and directional recommendation system for regional natural fund, which has the following beneficial effects:

[0047] 1. The regional natural fund's academic community mining and targeted recommendation system builds a multi-source heterogeneous relationship graph, integrating multi-dimensional information such as social relationships, research directions, and professional stratification to construct a three-layer heterogeneous graph of users, projects, and fields. This system deeply explores users' potential interests and cross-disciplinary needs, achieving precise matching of researchers and projects and intelligent recommendations. Furthermore, it dynamically adjusts recommendations based on user search keywords and browsing history, balancing personalized preferences with field expansion, improving recommendation accuracy and diversity, and ensuring efficient system operation and precise service for users.

[0048] 2. This system, developed for mining and targeting academic groups within the region's natural science foundation, targets cold-start scenarios. This system integrates user professional categories and behavioral data to quickly provide basic recommendations within the same field for new users. This effectively addresses the data sparsity issue inherent in traditional methods and improves the coverage and diversity of recommendations during the cold-start phase. Furthermore, through hierarchical mapping of professional disciplines, it breaks down barriers between sub-fields, enabling cross-sub-field recommendations and broadening the scope of user information acquisition, satisfying new users' need for cross-field information exploration and enhancing the richness and novelty of recommendation results.

[0049] 3. The regional natural fund's academic community mining and targeted recommendation system automates the entire process, from data file parsing to knowledge graph construction and visualization, reducing manual involvement and improving construction efficiency and quality. The Neo4j graph database stores entities, relationships, and attributes in the knowledge graph, ensuring efficient data storage and fast query. Using Echarts, complex knowledge graphs are presented in an intuitive and easy-to-understand graphical format, making it easier for users to understand and analyze relationships between data, providing strong support for scientific research analysis and decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0051] Figure 1 This is the main structural diagram of the system of the present invention;

[0052] Figure 2 Detailed architecture diagram of the system of the present invention;

[0053] Figure 3 This is a use case diagram for administrators of the present invention;

[0054] Figure 4 This is a diagram of the use cases used by researchers in the present invention;

[0055] Figure 5 This is a schematic diagram of a multiple mixed content recommendation module of the present invention;

[0056] Figure 6 Schematic diagram of the knowledge graph module of the present invention. DETAILED DESCRIPTION

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0058] This application embodiment provides a system for mining and recommending academic groups for regional natural fund projects. This system addresses the challenges faced by traditional methods of managing researchers and projects, which rely on fixed processes and standards and struggle to accurately capture the individual needs of researchers, projects, and academic groups as the number of researchers and projects grows and academic groups emerge. In project application, execution, conclusion, and outcome evaluation, there are issues such as insufficient information integration and analysis, inflexible resource allocation, and incomplete outcome evaluation.

[0059] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0060] Example 1: The present invention discloses a system for mining and recommending academic groups of regional natural science funds. Figure 1-6 As shown in the figure, the system is a highly integrated and fully functional comprehensive platform designed to optimize the allocation and management of regional natural funds through intelligent means, and to improve scientific research efficiency and innovation. The system consists of the following five key modules, which work together to achieve efficient operation and precise service:

[0061] 1. Data Management System

[0062] The data management system is the basic supporting module of the entire recommendation system, responsible for comprehensive and refined management of massive data related to researchers and research projects. This module has the following core functions:

[0063] 1. Data storage and maintenance: Use a high-performance database management system to store and maintain scientific research personnel and project data safely and efficiently.

[0064] 2. Data Query and Retrieval: This powerful data query feature allows researchers and administrators to quickly retrieve required researcher information or project data through keyword fuzzy searches. Query results are displayed in an intuitive table format, making it easy for users to quickly access key information.

[0065] 2. Knowledge Graph Module

[0066] The knowledge graph module is an efficient knowledge graph automatic construction and visualization system, mainly composed of a graph construction module and a front-end display module. It aims to quickly and accurately construct a knowledge graph and present it to users in an intuitive way. This module has the following core functions:

[0067] 1. Knowledge graph construction: Through steps such as data import and attribute mapping, raw data is converted into a knowledge graph. This supports the extraction of entities and relationships from structured data.

[0068] 2. Data storage and management: Use the Neo4j graph database to store entities, relationships, and attributes in the knowledge graph to ensure efficient data storage and fast query.

[0069] 3. Visual display: Based on the Echarts relationship diagram, the knowledge graph is displayed in an intuitive graphical form, allowing users to view graph details through operations such as zooming and dragging.

[0070] 3. Hybrid Recommendation Algorithm Module

[0071] This system provides a lightweight recommendation method and system based on multi-source heterogeneous relationship graphs. By integrating multi-dimensional information such as social relationships, research directions, and professional hierarchies, it builds an efficient recommendation model to solve the technical difficulties of traditional recommendation systems in heterogeneous data processing, cold start scenarios, and resource deployment.

[0072] 1. Multi-source heterogeneous relationship modeling method

[0073] 1.1 Build a three-layer relationship diagram of user-social-domain

[0074] Team Collaboration Network: Based on the airps_headid_memberid table, we extract the associations between team leaders and team members, adding edges with type='team' to the graph to accurately depict direct social relationships. For example, user A, as the project leader, establishes team edges with members B and C, forming a collaboration subgraph.

[0075] Research Direction: Community: Extracting relationships between team leaders and members based on the airps_headid_memberid table, adding edges with type = 'team' to the graph to accurately depict direct social relationships. For example, user A, as the project leader, establishes team edges with members B and C, forming a collaborative subgraph.

[0076] Professional Hierarchical Mapping: Use the text files in the Professional_classification directory to establish a mapping relationship between specific majors and major categories (e.g., the build_major_category_mapping function). This links user research interests to upper-level professional categories, forming a hierarchical graph structure that breaks down barriers between specific fields and supports cross-field recommendations.

[0077] 1.2 Heterogeneous Graph Node and Edge Types

[0078] The graph structure contains three types of nodes: users, projects, and professional categories, as well as three types of edges: team collaboration, research direction, and professional affiliation, comprehensively depicting the complex relationships between users, users, projects, and projects.

[0079] 2. Hybrid recommendation strategy

[0080] 2.1 Cold Start Scenario Processing

[0081] Professional field recommendations: Get the user's professional category through get_user_category, retrieve projects responsible for users in the same category based on the recommend_projects_by_category function, and quickly provide popular or representative projects in the field.

[0082] Fusion of search and browsing history: Combining user search keywords (search_projects function) and browsing history (get_viewed_projects_ids), the content recommender (ContentBasedRecommender) is used to calculate project similarity, supplement personalized recommendations, and improve diversity in the cold start phase.

[0083] 2.2 Similarity calculation and result fusion

[0084] Number of common neighbors: nx.common_neighbors is used to calculate the number of common connections between users and their neighbors, which measures the degree of social closeness.

[0085] Research direction matching: Users with the same research direction assign higher similarity weights.

[0086] Professional field expansion: add users who are in the same professional category but not directly related, expand the recommendation scope, and avoid "field islands".

[0087] 3. Lightweight Graph Recommendation Framework

[0088] 3.1 Efficient Implementation Based on Networkx

[0089] The lightweight graph library NetworkX is used to build graph structures, eliminating the need for complex distributed computing frameworks and supporting fast processing of small and medium-sized data. The core steps include:

[0090] Data loading: Read user, team, and project data from the database through the load_data function and build a user_projects mapping (a list of user IDs to their responsible project IDs).

[0091] Graph construction: The build_user_graph function adds nodes and edges in batches, with a time complexity of O(N 2 )(N is the number of users with the same research direction) to ensure real-time performance.

[0092] Similarity calculation: In recommend_similar_users, local neighbor search and regularized scoring are used to reduce computational overhead.

[0093] 4. AI Robot Module

[0094] The AI ​​robot module is the artificial intelligence interaction core of the system, providing users with intelligent and automated interactive services, improving user experience and system response efficiency. Its main functions include:

[0095] 1. Response Generation: Using iFlytek's Spark 4.0 Ultra model, we can generate high-quality responses to user messages. This model excels in text generation, language understanding, knowledge question answering, logical reasoning, and mathematical skills, providing accurate and diverse responses.

[0096] 2. Contextual Memory: Supports contextual memory and can understand the context of the user's conversation. This allows the robot to maintain coherence and consistency in multiple rounds of conversation, providing a more natural and accurate interactive experience.

[0097] 3. Intelligent Project Search and Recommendation: Based on the user's chat content, intelligently search and recommend related projects. This function analyzes the user's conversation content, identifies keywords in the conversation, and combines it with the system's project database to provide users with accurate project recommendations.

[0098] 5. User Interface

[0099] The user interface is an important part of the direct contact between the system and the user. Its design style is concise and clear, and the functional layout is clear and reasonable. It fully considers the operating habits and usage needs of scientific researchers and administrators, ensuring that users can use the system conveniently and efficiently. Its main features are as follows:

[0100] 1. Simple and intuitive interface design: Adopting a modern interface design style, the app features a concise layout, clear color schemes, and intuitive iconography, providing users with a comfortable and user-friendly visual experience. The layout of interface elements follows user operation logic, placing commonly used functional modules in prominent locations to reduce user operation steps and improve operational efficiency.

[0101] 2. Clearly defined functional modules: The system's main functional modules are clearly divided, and a clear navigation bar or menu guides users to quickly access the required functional modules. The operation process of each functional module is simple and easy to understand, and detailed prompts are provided to help users get started quickly.

[0102] In summary, the academic group mining and targeted recommendation system of the regional natural fund achieves accurate mining, intelligent recommendation and efficient management of scientific researchers and projects through the coordinated operation of the data management system, knowledge graph module, hybrid recommendation algorithm module, AI robot module and user interaction interface, providing strong technical support and decision-making basis for the rational allocation of regional natural funds and scientific research innovation.

[0103] The following is a detailed description of the advantages of each part of this software:

[0104] 1. Django framework and ORM mechanism

[0105] 1. High development efficiency: The ORM mechanism built into the Django framework allows developers to use Python classes and objects to operate the database without having to manually write SQL statements, which reduces learning costs and improves development efficiency.

[0106] For example, querying all records in the user table only requires this line of code:

[0107] Person.objects.all()

[0108] 2. Strong database compatibility: Supports multiple database backends, such as PostgreSQL, MySQL, SQLite, etc. Developers can switch between different databases without rewriting a lot of code.

[0109] 3. Strong security: ORM automatically redirects user input, reducing the risk of security vulnerabilities such as SQL injection. At the same time, the Django framework itself also provides many security features, such as SQL injection protection and CSRF protection, allowing developers to easily build secure and reliable web applications.

[0110] 4. Scalability and modularity: Django's architecture follows the loosely coupled design concept and adopts a modular development approach. Each of its components can be used or replaced independently, and its flexibility and scalability are very high.

[0111] 2. Knowledge Graph Module

[0112] 1. Fast Parsing and Importing: This platform can quickly parse large amounts of data files and batch import them into graph databases to build knowledge graphs, effectively improving the efficiency of knowledge graph construction. Data in CSV files is parsed and processed using the pandas library, an open-source data analysis library that efficiently processes structured data, such as filtering, merging, and transforming data. This facilitates the extraction of node and relationship information required for knowledge graph construction from raw data files.

[0113] 2. Graph data processing technology: Use the py2neo library to interact with the Neo4j database. py2neo provides a series of interfaces and classes that can easily operate nodes, relationships, graphs and other elements in Neo4j, and can implement operations such as creation, query, update, and deletion of nodes and relationships.

[0114] 3. Reduce manual intervention: It reduces manual intervention and the error rate of manually constructing knowledge graphs, while improving the stability and repeatability of the construction process.

[0115] 4. Front-end visualization technology: Echarts can be used to display complex knowledge graphs in an intuitive and easy-to-understand graphical form, making it easier for users to understand and analyze the relationship between data and provide strong support for decision-making.

[0116] 3. Hybrid Recommendation Algorithm Module

[0117] 1. Multi-dimensional heterogeneous relationship modeling: Integrate three types of information: social relationships, research direction communities, and professional hierarchies to construct a three-layer heterogeneous graph of users, projects, and fields. This breaks the limitations of a single interaction graph and deeply captures users' potential interests and cross-domain related needs.

[0118] 2. Cold start scenario adaptation: By integrating user professional categories and behavioral data (search / browsing history), new users without any interaction can receive basic recommendations in the same field. This solves the data sparsity problem of traditional methods and improves the coverage and diversity of recommendations during the cold start phase.

[0119] 3. Lightweight framework design: Graph processing is implemented based on NetworkX, reducing reliance on computing power, supporting efficient deployment of small and medium-sized data, and providing fast response speeds to meet real-time recommendation needs.

[0120] 4. Hybrid recommendation strategy: Dynamically combines content similarity (TF-IDF + cosine similarity) and graph structure information (social relationships, research direction matching), taking into account personalized preferences and field expansion, to improve recommendation accuracy and diversity.

[0121] 5. Modularity and scalability: The code structure is decoupled (image recommender, content recommender, tool module), supports independent expansion and function iteration, is compatible with data adaptation in different fields, and reduces maintenance and secondary development costs.

[0122] 6. Cross-domain knowledge integration: Break down the barriers between sub-domains through professional layered mapping, achieve cross-sub-domain recommendations, broaden the scope of user information acquisition, and enhance the richness and novelty of recommendation results.

[0123] 4. AI Robot Module

[0124] 1. Automatically generate reply content: Using the Spark4.0_ultra large model, it can generate high-quality reply content for messages sent by users, providing users with accurate and useful information.

[0125] 2. Contextual understanding: Supports contextual memory and can understand the contextual information of user conversations, making the conversations more coherent and natural, and better meeting the user's consulting needs.

[0126] 3. Intelligent search recommendation: The intelligent project search algorithm analyzes user chat content, processes and segments the chat content, automatically extracts keywords, and uses the keywords to query the data management system, thereby recommending projects related to the user's chat content. This further expands the user's channels for obtaining information and improves the system's intelligence and practicality.

[0127] 5. User Interface

[0128] 1. Simple and clear user interface design: The user interface design is simple, clear and reasonable, allowing users to quickly access various functions in the system, reducing the user's learning cost and improving the system's usability.

[0129] Improve operational efficiency: Reasonable interface layout and functional organization enable users to operate more efficiently. Whether querying and recommending projects and personnel, or interacting with AI robots, users can quickly find the entrance and complete the corresponding operations.

[0130] (1) The main structure of a Django-based project

[0131] 1. Project Directory (AI_research_paper_system)

[0132] The project directory is used to store files related to the core settings of the project. The following are the core files:

[0133] settings.py: It is the configuration file of the project, which includes many settings items such as static file configuration, media file configuration, database configuration, application configuration, middleware configuration, template configuration, etc.

[0134] urls.py: This is the URL routing configuration file for the project. It defines the mapping between the project's URL patterns and view functions. URL patterns are associated with view functions through functions like path(), thereby distributing user requests.

[0135] wsgi.py: Used to support the Web Server Gateway Interface (WSGI), which is the interface between the Django project and WSGI-compatible web servers. It enables Django projects to be deployed on servers that support WSGI, such as Gunicorn, uWSGI, etc.

[0136] 2. Application Catalog (AIRPS)

[0137] The application directory (AIRPS) is an independent functional module of the project. The application directory encapsulates a group of related functions, making it easy to modularize the development of Django projects and improve the maintainability and reusability of the code. The following is an introduction to the functions of each file and directory in this directory:

[0138] middleware: This directory stores the application's middleware, such as auth.py, which controls the processing of user authentication information.

[0139] Migrations: This directory stores model migration files. When a model class changes, Django generates a migration file to record the changes. It then applies these changes to the database by running the migration command, updating the database structure. Migration files are key to database version control, making it easy to manage changes to the database structure during collaborative development.

[0140] templates: This directory stores application template files. Template files are HTML files that can contain template variables and template tags, which are used to dynamically generate HTML content. View functions use the template rendering engine to combine template files with data to generate the final HTML response and send it to the user.

[0141] utils: This directory stores various utility tools within the application. These include the AI ​​robot, hybrid algorithm recommendation module, paging module, and file upload module. These modules follow a modular design and provide specific functionality for the software.

[0142] Views: This directory contains view files, which handle HTTP requests and return responses. A view function receives a request object as a parameter, performs logical processing based on the request content, and then returns a response object. View functions can call model classes to perform database operations or render templates to generate HTML responses.

[0143] admin.py: Used to register models with the Django admin backend. By registering models in admin.py, the model data can be managed in the Django admin backend, making it easier for developers to add, delete, modify, and query data.

[0144] models.py: defines the data model of the application and is the core part of Django ORM (Object Relational Mapping). The structure of the database table is described by defining model classes. Each model class corresponds to a table in the database, and the attributes of the model class correspond to the fields in the table. For example:

[0145] class Person(models.Model):

[0146] """Scientific researchers"""

[0147] id=models.CharField(verbose_name='id',max_length=50,primary_key= True)

[0148] name=models.CharField(verbose_name='name',max_length=50)

[0149] sex = models.CharField(verbose_name = 'sex', max_length = 50)

[0150] spusername = models.CharField(verbose_name = 'Approval Unit', max_length = 50)

[0151] xuew = models.CharField(verbose_name = 'degree', max_length = 50)

[0152] zhic = models.CharField(verbose_name = 'Title', max_length = 50)

[0153] dwmc = models.CharField(verbose_name = 'Unit Name', max_length = 50)

[0154] sxzy = models.CharField(verbose_name = 'Major', max_length = 50)

[0155] stae = models.CharField(verbose_name = '1: person in charge; 0: team member', max_length ＝50)

[0156] sbnd = models.CharField(verbose_name = 'Declaration Year', max_length = 50)

[0157] birthday=models.CharField(verbose_name='Birthdate',max_length=50)

[0158] starttime=models.CharField(verbose_name='start time',max_length=50)

[0159] endtime = models.CharField(verbose_name = 'End time', max_length = 50)

[0160] password=models.CharField(verbose_name='password',max_length=50, default='1')

[0161] This class defines a model of a researcher, corresponding to the table storing researchers in the database.

[0162] tests.py: This file stores your application's test code. Django provides a powerful testing framework, allowing you to write test cases to test your application's functionality, including unit tests and integration tests. These tests can check that view function return values ​​and model class method behavior meet expectations, helping to ensure code quality and reliability.

[0163] 3. Static files (static): Stores static resources for your project, such as CSS files, JavaScript files, and images. Django provides a static file management mechanism, allowing you to configure the path and access method for static files through settings such as STATIC_URL and STATICFILES_DIRS. During development, static files can be served directly from the file system. In a production environment, static files are typically collected in a designated directory and then served by a dedicated static file server (such as nginx).

[0164] 4. MySQL Database: MySQL is an open-source relational database management system. It uses SQL (Structured Query Language) to manage data and offers advantages such as high performance, high reliability, ease of use, and low cost. MySQL supports multiple operating systems, including Linux and Windows, and is widely used in various applications, including websites and enterprise applications. It efficiently stores, retrieves, and manages large amounts of data, providing users with flexible data manipulation and management capabilities.

[0165] 5. Neo4j graph database: Neo4j is a high-performance graph database that stores structured data on the network rather than in tables. Neo4j provides powerful graph data storage and query capabilities, can efficiently handle complex relational data, support transactional operations, and ensure data consistency and integrity. It has a flexible data model that can easily expand and modify data structures. It is suitable for application scenarios that need to handle complex relationships, such as social networks, recommendation systems, and knowledge graphs, helping users quickly discover connections and patterns between data. Django uses the Py2neo library to communicate with the neo4j database. Py2neo is a library for Python specifically designed to interact with the Neo4j graph database. It provides a simple and easy-to-use interface that allows developers to easily connect to the Neo4j database through Python code and perform operations such as adding, deleting, modifying, and querying graph data. Py2neo supports transactions to ensure the atomicity of data operations, and also provides a rich API to operate basic elements of graph databases such as nodes, relationships, and paths, allowing developers to efficiently build applications based on graph databases.

[0166] (2) Data management system based on DjangoORM mechanism

[0167] DjangoORM (Object-Relational Mapping) is one of the core components of the Django framework. It allows developers to use Python code to operate the database without writing SQL statements. By mapping database tables to Python classes (models), DjangoORM provides an intuitive and efficient way to manage data in the database.

[0168] 1. Model definition

[0169] Each database table corresponds to a Python class (model), which inherits from django.db.models.Model. The attributes in the model class correspond to the fields in the table. Django provides a variety of field types (such as CharField, IntegerField, DateField, etc.) to define the data type and constraints of the field.

[0170] 2. Create data

[0171] DjangoORM mechanism inserts data through the save() method of the model class or the objects.create() method.

[0172] 3. Delete data

[0173] Django ORM mechanism deletes data through the delete() method of the model instance or the delete() method of the query set.

[0174] 4. Update data

[0175] After the DjangoORM mechanism obtains the object through the query, it modifies the attributes and calls the save() method, or uses the update() method to update in batches.

[0176] 5. Query Data

[0177] DjangoORM mechanism uses objects query set (QuerySet) for flexible query operations, supporting filtering, sorting, paging, etc.

[0178] 6. Security

[0179] DjangoORM processes SQL queries internally, automatically escaping parameters to effectively prevent SQL injection attacks. It supports transaction operations to ensure data consistency and integrity.

[0180] (3) Multiple algorithm hybrid recommendation system

[0181] 1. Build a three-layer relationship diagram of user-social-domain

[0182] Team Collaboration Network: Based on the airps_headid_memberid table, we extract the associations between team leaders and team members, adding edges with type='team' to the graph to accurately depict direct social relationships. For example, user A, as the project leader, establishes team edges with members B and C, forming a collaboration subgraph.

[0183] Research Direction: Community: Extracting relationships between team leaders and members based on the airps_headid_memberid table, adding edges with type = 'team' to the graph to accurately depict direct social relationships. For example, user A, as the project leader, establishes team edges with members B and C, forming a collaborative subgraph.

[0184] Professional Hierarchical Mapping: Use the text files in the Professional_classification directory to establish a mapping relationship between specific majors and major categories (e.g., the build_major_category_mapping function). This links user research interests to upper-level professional categories, forming a hierarchical graph structure that breaks down barriers between specific fields and supports cross-field recommendations.

[0185] 2. Hybrid recommendation strategy

[0186] Cold start scenario processing: Use get_user_category to obtain the user's professional category, use the recommend_projects_by_category function to retrieve projects responsible for users in the same category, and quickly provide popular or representative projects in the field.

[0187] The source code is as follows:

[0188]

[0189]

[0190]

[0191] Search and browsing history integration: Combine user search keywords (search_projects function) and browsing history (get_viewed_projects_ids), calculate project similarity through the content recommender (ContentBasedRecommender), supplement personalized recommendations, and improve diversity in the cold start phase. The source code is as follows:

[0192]

[0193]

[0194]

[0195] 3. Lightweight graph recommendation framework Networkx

[0196] The lightweight graph library NetworkX is used to build graph structures, eliminating the need for complex distributed computing frameworks and supporting fast processing of small and medium-sized data. The core steps include:

[0197] Data loading: Read user, team, and project data from the database through the load_data function and build a user_projects mapping (a list of user IDs to their responsible project IDs).

[0198] Graph construction: The build_user_graph function adds nodes and edges in batches, with a time complexity of O(N 2 )(N is the number of users with the same research direction) to ensure real-time performance.

[0199] Similarity calculation: In recommend_similar_users, local neighbor search and regularized scoring are used to reduce computational overhead.

[0200] (4) AI Robot Module

[0201] 1. Operation logic

[0202] The main view function of the AI ​​robot is as follows:

[0203]

[0204]

[0205] The AI ​​bot receives messages from the frontend. It first sends the message to the Spark-4.0-ultra chat class, generates a reply, and temporarily stores the chat context. The AI ​​bot then sends the message to the keyword automatic recognition and search module for automatic word segmentation, ultimately returning up to three relevant projects. If any projects are returned, they are appended to the chat reply. Otherwise, a "No related projects recommended" message will be added.

[0206] 2. Spark-4.0-ultra large model chat class

[0207] The source code of the Spark-4.0-ultra large model chat class is as follows:

[0208]

[0209]

[0210] This chat class can generate high-quality chat replies and record context.

[0211] 3. Keyword automatic recognition and search module

[0212] The source code of the keyword automatic recognition search module is as follows:

[0213]

[0214]

[0215]

[0216] The keyword automatic recognition and search function can segment chat text, automatically identify keywords in the chat content, and use the Django ORM mechanism to search for items, returning up to a specified number of item search contents.

[0217] (5) Knowledge Graph Module

[0218] 1. Innovative hybrid construction model: This model combines data analysis with graph construction, improves data import efficiency through batch processing technology, and adopts strategies such as dynamic adjustment of model weights to improve the accuracy and adaptability of graph construction.

[0219] 2. Unique graph construction process: including data parsing, node relationship creation, batch transaction submission, graph data export, and other links, as well as automated processing of different types of data and relationships throughout the process.

[0220] 3. Efficient automated construction method: Achieve full-process automation from data file parsing to knowledge graph construction and visualization, reduce manual participation, and improve construction efficiency and quality.

[0221] The complete source code of the knowledge graph construction part is as follows:

[0222]

[0223]

[0224]

[0225] 4. Efficiently and comprehensively build knowledge graphs

[0226] The knowledge graph module can automatically construct a knowledge graph based on the researcher's ID, showing the relationship between the specified researcher and other researchers and projects. The following is the source code:

[0227]

[0228]

[0229]

[0230]

[0231]

[0232] This function mines relevant information about target researchers from multiple dimensions by querying collaborative relationships, shared expertise, and related projects, and integrates this information into a knowledge graph. Furthermore, the function performs deduplication after each added node and relationship, ensuring data accuracy and consistency. This multi-dimensional query and data deduplication design enables the function to efficiently generate a comprehensive and accurate knowledge graph, providing strong support for subsequent scientific research analysis and decision-making.

[0233] (1) Multiple Algorithm Hybrid Recommendation System

[0234] 1. Multi-dimensional heterogeneous relationship modeling and cross-domain knowledge integration

[0235] By integrating social relationships, research community, and professional hierarchies, we construct a three-layer heterogeneous graph of users, projects, and domains, breaking the limitations of a single interaction graph. Furthermore, by leveraging professional hierarchical mapping, we break down the barriers between sub-domains and enable cross-sub-domain recommendations. This enables the recommendation system to deeply capture users' potential interests and cross-domain related needs, broadening the scope of user information acquisition and enhancing the richness and novelty of recommendation results.

[0236] 2. Use lightweight networks and efficient deployment

[0237] Using NetworkX for graph processing and a lightweight framework, we significantly reduce reliance on computing power, enabling the system to efficiently process small and medium-sized data and enabling rapid deployment. This lightweight network design not only delivers fast response times but also meets real-time recommendation requirements, providing users with timely and accurate scientific research project recommendations and enhancing the user experience.

[0238] 3. Hybrid recommendation strategy and modular scalability

[0239] Dynamically combining content similarity (TF-IDF + cosine similarity) with graph structure information (social connections, research direction matching), it balances personalized preferences with domain expansion, improving recommendation accuracy and diversity. The decoupled code structure (graph recommender, content recommender, and tool modules) supports independent expansion and functional iteration, is compatible with data adaptation in different fields, and reduces maintenance and secondary development costs.

[0240] (2) AI robot based on Spark-4.0-ultra large model and keyword automatic recognition and search

[0241] 1. Multimodal Interaction and Context Management

[0242] The AI ​​bot is capable of processing information in both natural language conversation and project search modes. After receiving a user's message, it generates a natural language response using the Spark-4.0-ultra large model and searches for relevant projects using an automatic keyword recognition search module. This multimodal interaction not only enables fluent conversations but also provides detailed project information based on the conversation, enriching the interaction. Furthermore, the bot records the chat context during communication with the user, enabling it to understand the coherence and context of the conversation, thereby generating responses that better align with the user's intent and improving the user experience.

[0243] 2. Efficient and stable dialogue generation and context recording

[0244] The Spark-4.0-ultra large-model chat class features efficient conversation generation capabilities, enabling rapid generation of high-quality chat responses. It records context by preserving conversation history, allowing the bot to reference previous exchanges in subsequent conversations and better understand the user's intentions and needs. This contextual recording mechanism is crucial for handling complex conversation scenarios, enabling the bot to provide more coherent and personalized services, and enhancing interactivity and trust between users and the bot.

[0245] 3. Automatic reconnection and error handling mechanism

[0246] If errors occur during conversation generation, such as network issues or model anomalies, the chat class automatically reconnects and attempts to generate a response again. This automatic reconnection and error handling mechanism effectively avoids conversation interruptions caused by temporary failures, ensuring that the robot can continue to provide stable service to users. This improves system reliability and stability, reduces user inconvenience, and enhances the robot's usability in real-world applications.

[0247] 4. Accurate keyword extraction and project search

[0248] The automatic keyword recognition and search module segments user chat text and automatically identifies keywords. These keywords are then used to search for items in the database through the Django ORM mechanism. This keyword-based search method quickly locates items related to the user's chat content, improving the efficiency and accuracy of information retrieval. It enables the bot to provide relevant project information in a timely manner based on the user's conversation content, enhancing the bot's information provision capabilities and practicality.

[0249] 5. Multi-field search and result optimization

[0250] When searching for projects, the module searches multiple fields (such as Chinese keywords, project names, first-level disciplines, third-level disciplines, project summaries, etc.) in a certain order. This multi-field search strategy can more comprehensively cover project information and increase the possibility of finding relevant projects. At the same time, the search process will be optimized according to the set search quantity limit, and the search will stop when a sufficient number of relevant projects are found. This avoids returning too much irrelevant information, improves the quality and practicality of search results, and enables the robot to provide users with more accurate and valuable project recommendations.

[0251] (3) Rapid construction of the researcher-project knowledge graph and front-end display module based on the Neo4j graph database

[0252] 1. Combining data analysis and graph construction

[0253] Using data parsing as a precursor to graph construction ensures the accuracy and consistency of input data, laying a solid foundation for subsequent graph construction. Furthermore, during the construction process, by dynamically adjusting model weights and other strategies, we can flexibly optimize the accuracy and adaptability of graph construction based on different data characteristics and business needs, enabling the knowledge graph to better reflect actual business logic and knowledge relationships.

[0254] 2. Full process automation

[0255] The entire process from data file parsing to knowledge graph construction and visualization is automated, greatly reducing manual intervention. For example, in the provided code, from parsing the data file, initializing the graph builder, clearing the graph, creating various nodes and relationships, and finally exporting the data to the Echarts format, the entire process does not require excessive manual intervention, saving a lot of time and labor costs, and improving construction efficiency and quality.

[0256] 3.Automated processing of different types of data and relationships

[0257] Throughout the entire build process, different types of data and relationships can be automatically processed without manual intervention. This not only improves build efficiency but also reduces the risk of human error, ensuring the quality and consistency of the knowledge graph. For example, in the code, through simple loops and conditional judgments, information such as people, teams, and projects can be automatically extracted from the data and the corresponding nodes and relationships can be created, greatly simplifying the build process.

[0258] 4. Multi-dimensional information mining

[0259] Taking the researcher's ID as the starting point, it is possible to mine relevant information about the target researcher from multiple dimensions such as collaborative relationships, relationships with the same profession, and related projects, and integrate this information into a knowledge graph. This multi-dimensional query method can comprehensively display the business connections and knowledge connections of researchers, providing richer and more comprehensive data support for scientific research analysis and decision-making. For example, in the "id_random_graph" function, different Cypher query statements are used to query collaborative relationships, relationships with the same profession, and related projects, respectively, and obtain nodes and relationships related to the target researcher from multiple perspectives. Deduplication is performed after each addition of nodes and relationships to ensure the accuracy and consistency of the data, avoid duplicate nodes and relationships in the knowledge graph, and improve the quality and reliability of the knowledge graph. This is very important for various subsequent analyses and applications based on the knowledge graph, and can ensure the accuracy and credibility of the analysis results.

[0260] 5. Comprehensively display knowledge connections

[0261] The constructed knowledge graph not only includes direct relationships between scientific researchers, such as collaborative relationships, relationships with the same profession, etc., but also displays indirect relationships between scientific researchers and projects, as well as indirect connections between scientific researchers through information such as related projects. This enables the knowledge graph to more comprehensively reflect the knowledge network and business connections of scientific researchers, and provides strong support for cooperation, project recommendations, knowledge sharing, etc. among scientific researchers.

[0262] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0263] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A regional natural fund academic group mining and directional recommendation system, characterized by: include: Data management system, including: A database storage unit uses a relational database to store information; Multi-field joint search unit, which performs cross-field joint query in the information table and project information table, and returns the search results in table form; Knowledge graph module, including: Neo4j graph database storage unit, used to store graph structure data; Echarts visualization engine, used to generate researcher-project relationship maps; Hybrid recommendation algorithm module, including: The three-layer relationship graph construction unit is used to generate user-user team relationship edges based on the team collaboration table, generate user-user research relationship edges based on research directions, and generate user-professional category edges through professional classification mapping files; The cold start processing unit is used to search for projects in the same field according to their professional categories when the user has no historical interaction data, and integrate the real-time search keyword matching results; AI robot module, including: Large model interaction unit, used to parse user natural language requests and generate text responses; Keyword search unit, used to perform word segmentation on user input text and perform multi-field matching retrieval; User interface module.

2. The academic group mining and directional recommendation system for regional natural fund according to claim 1 is characterized in that: The database storage unit adopts a MySQL relational database, and the storage unit includes a scientific researcher table, a project information table, and a team leader-member relationship table.

3. The academic group mining and directional recommendation system for regional natural fund according to claim 1 is characterized in that: The multi-field joint retrieval unit supports performing OR logical joint queries on the name, gender, degree, title, unit name, and major fields of the scientific researcher table.

4. The academic group mining and directional recommendation system for regional natural fund according to claim 1 is characterized in that: The knowledge graph module includes: An automated graph construction unit that uses the pandas library to parse CSV-formatted researcher collaboration data files; The node relationship batch creation unit uses the py2neo library to batch import the parsed data into the Neo4j database.

5. The academic group mining and directional recommendation system for regional natural fund according to claim 4 is characterized in that: The node relationship batch creation unit includes: Create a sub-unit for the personnel collaboration relationship and map the relationship between the team leader and members to a "team leader-belongs" relationship edge; Create a sub-unit for the scientific research participation relationship and map the association between personnel and projects as a "participation" relationship edge.

6. The academic group mining and directional recommendation system for regional natural fund according to claim 1 is characterized in that: The three-layer relationship diagram construction unit includes: The team collaboration subgraph generation module constructs a relationship edge with type='team' based on the team leader ID and member ID fields; The professional hierarchical mapping module associates the user's research direction with the predefined professional category nodes through the professional classification mapping file.

7. The academic group mining and directional recommendation system for regional natural fund according to claim 1 is characterized in that: The cold start processing unit includes: The professional field search submodule executes SQL queries to obtain projects that users in the same professional category are responsible for; The behavioral data fusion submodule calls the TF-IDF vectorizer to calculate the cosine similarity between user search keywords and project summaries.

8. The academic group mining and directional recommendation system for regional natural fund according to claim 1 is characterized in that: The AI ​​robot module includes: Context memory storage, used to save the context state of the user's continuous conversation; The multi-field search executor performs keyword matching in the priority order of Chinese keywords → project name → first-level discipline → third-level discipline → project summary.

9. The academic group mining and directional recommendation system for regional natural fund according to claim 1 is characterized in that: The user interaction interface module is constructed using the Django framework, and the interface module includes: The scientific research personnel management function area provides an interface for adding, deleting, modifying and checking personnel information; Knowledge graph display area, dynamically loading the scientific research relationship network diagram generated by Echarts; The recommendation result display area displays a list of recommended items based on graph algorithms and content matching in columns.

10. The academic group mining and directional recommendation system for regional natural fund according to claim 1 is characterized in that: The system deployment architecture includes: Django application server, used to process HTTP requests and call various functional modules; Parallel MySQL database and Neo4j graph database, storing structured data and graph relational data respectively; NetworkX graph computing engine, used to process user similarity calculation requests in real time.

Images