A two-dimensional magnetic material use driven reverse recommendation method and system

Abstract

The application discloses a two-dimensional magnetic material use driving reverse recommendation method and system, belongs to the cross application field of material informatics and artificial intelligence technology, and the method comprises the following steps: acquiring patent data and attribute data corresponding to two-dimensional magnetic materials by using a multi-source data fusion module, and obtaining a multi-source fusion database through an ETL data stream integration processing mechanism; receiving a use requirement input by a user by using a multi-AI collaborative intelligent reasoning module, intelligently reasoning the use requirement, and generating a recommendation result; realizing a multi-key polling technology by using an API polling mechanism of a high-availability system architecture module, and constructing a multi-level fault-tolerant processing system; providing a responsive Web interface to support the user to input the use requirement by using an intelligent user interaction module, determining the use correlation, and displaying attribute combination prediction results and intelligent recommendation results. The application solves the problems of data fragmentation, unstructured semantic barriers and single recommendation mode in the existing material screening method.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary application field of materials informatics and artificial intelligence technology, specifically involving a method and system for reverse recommendation driven by the application of two-dimensional magnetic materials. Background Technology

[0002] Currently, two-dimensional magnetic materials, as novel functional materials, demonstrate significant research and application value in areas such as electronic devices, information storage, and quantum computing. However, researchers still face numerous challenges in the material screening process, mainly in the following aspects:

[0003] Data fragmentation: Existing mainstream materials databases such as Materials Project only provide structured data on physical properties, lacking application-related information, making it difficult to support application-oriented material screening;

[0004] Unstructured semantic barrier: A large amount of knowledge about the actual use of materials exists in patent literature, expressed in natural language, which makes it difficult to directly extract and apply in a structured way;

[0005] The recommended approach is limited: most existing tools are forward-looking material lookup tools, which cannot meet the reverse exploration needs of finding materials "starting from application".

[0006] Therefore, building a novel system that integrates structured data of material properties with semantic information about their uses to support reverse recommendation has become an important path to promote the discovery of new materials. Summary of the Invention

[0007] To address the aforementioned shortcomings in existing technologies, this invention provides a two-dimensional magnetic material application-driven reverse recommendation method and system that solves the problems of data fragmentation, unstructured semantic barriers, and a single recommendation method in existing material screening methods.

[0008] To achieve the above-mentioned objectives, the technical solution adopted by this invention is: a method for reverse recommendation driven by the application of two-dimensional magnetic materials, comprising the following steps:

[0009] S1: Use the multi-source data fusion construction module to obtain the patent data and attribute data corresponding to the two-dimensional magnetic materials, construct a dual-table database, and obtain the multi-source fused database through the ETL data stream integration processing mechanism;

[0010] S2: Utilize a multi-AI collaborative intelligent reasoning module to receive user input regarding usage requirements, perform intelligent reasoning on them, and generate recommendation results;

[0011] S3: Utilizes a high-availability system architecture module to implement multi-key polling technology using an API polling mechanism, constructs a multi-layered fault-tolerant processing system, and achieves database query optimization and asynchronous front-end loading through a performance optimization engine;

[0012] S4: Utilize the intelligent user interaction module to provide a responsive web interface that supports user input of usage requirements, determines usage relevance, displays attribute combination prediction results and intelligent recommendation results, and provides a visual representation of the results.

[0013] Further, step S1 includes the following sub-steps:

[0014] S11: The intelligent annotation submodule based on patent data retrieves corresponding patent information from Google Patents Public Datasets by using SQL technology to query based on keywords related to two-dimensional magnetic materials. It then uses the Tongyi Qianwen big model and Few-shot prompting engineering technology to generate intelligent tags for the patent abstract.

[0015] S12: The automatic calculation and classification submodule based on material properties obtains the magnetic information of two-dimensional materials from the Materials Project and the electrical information of two-dimensional materials from the 2DMatPedia database.

[0016] S13: Through the material property calculation engine, the material type is automatically determined to be metal, semiconductor or insulator based on the band gap value, and ferromagnetism or antiferromagnetism is automatically identified through the magnetic sorting parameter.

[0017] S14: Construct a dual-table structure of materials_magnetic and application tables, and establish an MPID standardized index;

[0018] S15: Data cleaning is performed based on the ETL data stream integration processing mechanism, and the association mapping between patent data and material data is realized through MySQL.

[0019] Furthermore, in step S13, the material type is automatically determined as a metal, semiconductor, or insulator based on the band gap value, specifically as follows:

[0020] A band gap value of 0 indicates that the material type is metal.

[0021] A band gap value of 0-3 eV indicates that the material type is a semiconductor.

[0022] If the band gap value is greater than 3eV, the material type is determined to be an insulator.

[0023] Furthermore, step S2 includes the following sub-steps:

[0024] S21: Based on the intelligent verification subsystem, the relevantness of the received application requirements is judged, and intelligent prompts or rejection processing is performed for inputs with confidence levels below the threshold.

[0025] S22: Extract relevant dimensional information from the verified usage requirements based on the structured information extraction subsystem;

[0026] S23: Based on the XGBoost prediction engine, receive the structured information extraction results and perform triple feature encoding to predict the material type and magnetic characteristics;

[0027] S24: The RAG-based intelligent recommendation generation subsystem performs precise matching queries based on the predicted material type and magnetic characteristics, uses a dynamic query algorithm to filter materials from the database, and generates a recommendation report that includes a description of the prediction results, an explanation of the technical principles, and relevant patent cases to support it.

[0028] Furthermore, in S22, the structured information extraction subsystem receives valid input confirmed by the intelligent verification subsystem, extracts application category, application subcategory, target material type and target magnetic type dimension information, adds semantic reasoning prompts for inputs of less than 50 characters, and provides default reasoning logic based on keywords.

[0029] Furthermore, the multi-level fault-tolerant processing subsystem in S3 includes full-process anomaly capture for input validation anomalies, model prediction anomalies, and database query anomalies, and is equipped with an intelligent degradation strategy.

[0030] The present invention also employs the following technical solution: a two-dimensional magnetic material application-driven reverse recommendation system, comprising:

[0031] The multi-source data fusion construction module includes a patent data intelligent annotation sub-module and a material property automatic calculation and classification sub-module, which are used to obtain patent data and property data corresponding to two-dimensional magnetic materials, construct a dual-table structure database, and obtain a multi-source fused database through an ETL data flow integration processing mechanism;

[0032] The multi-AI collaborative intelligent reasoning module includes an intelligent verification subsystem, a structured information extraction subsystem, an XGBoost prediction engine, and a RAG intelligent recommendation generation subsystem. It is used to receive user input regarding usage requirements, perform intelligent reasoning on them, and generate recommendation results.

[0033] The high availability system architecture module adopts an API polling mechanism to implement multi-key polling technology, builds a multi-layered fault-tolerant processing system, and achieves database query optimization and asynchronous front-end loading through a performance optimization engine.

[0034] The intelligent user interaction module provides a responsive web interface that supports user input of usage requirements, determines the relevance of usage, displays attribute combination prediction results and intelligent recommendation results, and provides a visual representation of the results.

[0035] The beneficial effects of this invention are:

[0036] High data coverage: The system integrates structured physical property data of approximately 440 two-dimensional materials and more than 4,000 patent use corpora, constructing a rich material-use mapping database;

[0037] Excellent recommendation accuracy: In the verification test, based on the application requirements input by the user, the overall matching accuracy of the system's application identification and material property combination prediction reached 82%;

[0038] Fast response time: In a local deployment environment, the average response time for a single recommendation request is less than 0.5 seconds, meeting the real-time interaction requirements of the online platform. Attached Figure Description

[0039] Figure 1 A flowchart illustrating a reverse recommendation method driven by the application of a two-dimensional magnetic material.

[0040] Figure 2 This is a sample diagram illustrating the text input by the user describing the application scenarios of magnetic materials.

[0041] Figure 3 This is a schematic diagram of the system's recommended results page.

[0042] Figure 4 This is a schematic diagram of the system's recommended results page 2.

[0043] Figure 5 The third illustration shows the system's recommended results page.

[0044] Figure 6 Show detailed property diagrams for the Materials Project.

[0045] Figure 7 This is a schematic diagram of the system's feedback interface for inputs used for unrelated purposes.

[0046] Figure 8 This is a schematic diagram of the system's feedback interface for inputs of unrelated purposes.

[0047] Figure 9 Three schematic diagrams illustrate the system's feedback interface for inputs used for unrelated purposes. Detailed Implementation

[0048] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0049] Example 1, as Figure 1 As shown, a reverse recommendation method driven by the application of two-dimensional magnetic materials includes the following steps:

[0050] S1: Use the multi-source data fusion construction module to obtain the patent data and attribute data corresponding to the two-dimensional magnetic materials, construct a dual-table database, and obtain the multi-source fused database through the ETL data stream integration processing mechanism;

[0051] S1 includes the following steps:

[0052] S11: The intelligent annotation submodule based on patent data retrieves corresponding patent information from Google Patents Public Datasets by using SQL technology to query based on keywords related to two-dimensional magnetic materials. It then uses the Tongyi Qianwen big model and Few-shot prompting engineering technology to generate intelligent tags for the patent abstract.

[0053] In this embodiment, the patent abstract is intelligently tagged by calling the Tongyi Qianwen Big Model and using Few-shot prompting engineering technology. The tag types are divided into 13 main tags (such as Electronics_and_Information_Storage), and each main tag contains 3 sub-tags (such as MRAM_STT, HDD_Read_Write_Head, Magnetic_Stripe_Card), constructing a refined application tag system of 39 types.

[0054] S12: The automatic calculation and classification submodule based on material properties obtains magnetic information (ferromagnetic, antiferromagnetic, etc.) of two-dimensional materials from the Materials Project and electrical information (conduction band, valence band, Fermi level, band gap, etc.) of two-dimensional materials from the 2DMatPedia database of the National University of Singapore.

[0055] S13: Through the material property calculation engine, the material type is automatically determined to be metal, semiconductor or insulator based on the band gap value, and ferromagnetism or antiferromagnetism is automatically identified through the magnetic sorting parameter.

[0056] Automatic classification algorithm:

[0057] if band_gap == 0: material_type = "metal"

[0058] elif 0 <band_gap ≤ 3: material_type = "semiconductor"

[0059] elif band_gap>3: material_type = "insulator"

[0060] In step S13, the material type is automatically determined as a metal, semiconductor, or insulator based on the band gap value. Specifically:

[0061] A band gap value of 0 indicates that the material type is metal.

[0062] A band gap value of 0-3 eV indicates that the material type is a semiconductor.

[0063] If the band gap value is greater than 3eV, the material type is determined to be an insulator.

[0064] S14: Construct a dual-table structure of materials_magnetic and application tables, and establish an MPID standardized index;

[0065] S15: Data cleaning is performed based on the ETL data stream integration processing mechanism, and the association mapping between patent data and material data is realized through MySQL;

[0066] Data cleaning: Design an iterative data cleaning algorithm to process incomplete, duplicate, and inconsistent original data through data comparison.

[0067] S2: Utilize a multi-AI collaborative intelligent reasoning module to receive user input regarding usage requirements, perform intelligent reasoning on them, and generate recommendation results;

[0068] S2 includes the following steps:

[0069] S21: Based on the intelligent verification subsystem, the relevantness of the received application requirements is judged, and intelligent prompts or rejection processing is performed for inputs with confidence levels below the threshold.

[0070] Further includes:

[0071] Collaboration mechanism: As the first line of defense in the entire reasoning chain, it provides high-quality input for subsequent modules;

[0072] Validation Algorithm: Based on the Tongyi Qianwen semantic understanding engine, specialized relevance judgment prompts are designed, specifically as follows:

[0073] Example: Basic Correlation Verification Process (Online Main Process)

[0074] Model and parameters

[0075] Model: qwen-turbo

[0076] Temperature: 0.0; top_p: 0.9; max_tokens: 256; Timeout: 30s

[0077] Return format: JSON (force response_format={"type":"json_object"})

[0078] Determination threshold

[0079] The confidence threshold θ = 0.70; the output is_valid ∈ {true,false}, confidence ∈ [0,1].

[0080] Judgment rule: If is_valid and confidence ≥ θ, then pass; otherwise, reject and provide a suggestion.

[0081] Prompt design

[0082] Key points for system prompts: The domain is limited to two-dimensional magnetic materials and applications; valid JSON output is required; positive and negative examples must be provided; emphasis is placed on semantic reasoning rather than keyword matching, even for short texts.

[0083] User prompt structure: Embedded user text + Judgment criteria + Output format requirements

[0084] Return format

[0085] {"is_valid": true / false, "confidence": 0.0-1.0, "reason": "brief reason"}

[0086] Code snippet (runnable)

[0087] import OpenAI from openai

[0088] SYSTEM_PROMPT = """You are an auditing expert in the field of materials science. Task: Determine whether user input is relevant to the application requirements of two-dimensional magnetic materials.

[0089] Only a JSON object must be output, with the fields: is_valid(bool), confidence(float), and reason(str).

[0090] Judgment criteria:

[0091] 1) Does it involve application areas such as materials science, electronic devices, sensors, storage, energy, medical, or machinery?

[0092] 2) Does it describe the functional requirements, performance requirements, or technical challenges?

[0093] 3) Is it possible to solve this problem through the properties of two-dimensional magnetic materials (magnetic order, magnetic response, magnetic coupling, etc.)?

[0094] 4) Exclude obviously irrelevant content (casual chat, subjects unrelated to the material, garbled text / malicious input).

[0095] Example:

[0096] Input: "Design a highly sensitive magnetic field sensor"

[0097] Output: {"is_valid": true, "confidence": 0.95, "reason": "Involves sensor and magnetic response"}

[0098] Confidence score: Output a confidence score of 0.0-1.0, with 0.7 set as the pass threshold;

[0099] Filtering mechanism: Intelligent suggestions or rejection are provided for inputs with confidence scores below a threshold;

[0100] Fault tolerance: When an API call fails, rule-driven backup verification logic is enabled, specifically:

[0101] When the large model validation interface fails to poll multiple keys and perform exponential backoff retries (up to 3 times) for 1 / 2 / 4 seconds, rule-driven backup logic is activated: the input is cleaned and segmented, and matched with domain keywords / patterns (such as "magnetic," "MRAM," "spin," "sensor," "magnetic storage," etc., co-occurring with words like "performance / requirement"). If a match is found, is_valid=true and confidence=0.6 is returned; otherwise, false and confidence=0.5 are returned. The time, key ID, number of retries, and judgment result are recorded for auditing and threshold optimization purposes.

[0102] S22: Extract relevant dimensional information from the verified usage requirements based on the structured information extraction subsystem;

[0103] The structured information extraction subsystem in S22 receives valid input confirmed by the intelligent verification subsystem, extracts application category, application subcategory, target material type and target magnetic type dimension information, adds semantic reasoning prompts for inputs of less than 50 characters, and provides default reasoning logic based on keywords.

[0104] Structured Information Extraction Subsystem – Specific Implementation Example (Reproducible)

[0105] Model and parameters

[0106] Model: qwen-turbo; temperature=0.0, top_p=0.9, max_tokens=256, timeout=30s

[0107] Return format: JSON (force response_format={"type":"json_object"})

[0108] Prompt design

[0109] The system is limited to outputting four fields: application_category, application_subcat, target_type, and target_mag; it includes an embedded 13x3 tag whitelist and value constraints; and it requires outputting only valid JSON.

[0110] user: Embed the original user text + judgment rules (must select the most matching first-level / second-level tags; type limited to semiconductor / metal / insulator; magnetic type limited to ferromagnetic / antiferromagnetic)

[0111] Short text adaptive

[0112] If the input length is less than 50 characters, the following prompt will appear: "Please determine the main scenario and technical purpose based on semantic reasoning, and do not perform keyword surface matching."

[0113] 50–200 characters, add "Please summarize the main idea before selecting tags".

[0114] Analysis and fallback logic

[0115] If any field is empty or null, a fallback rule is triggered: fill in the blanks based on keywords / semantic clues.

[0116] "Sensing / Detection / Stress / Encoding" → Sensing_and_Detection; Subclasses: containing "Stress / Deformation" → Magnetostrictive_Stress_Sensor; containing "Position / Encoding" → Position_Encoder; otherwise Leak_Magnetic_Test;

[0117] "Storage / MRAM / Magnetic Stripe Card" → Electronics_and_Information_Storage; Subclasses: containing "MRAM" → MRAM_STT; containing "Magnetic Stripe / Stripe" → Magnetic_Stripe_Card; otherwise HDD_Read_Write_Head;

[0118] "Relay / Switch / Actuation / Valve / Linear" → Actuation_and_Drive_Systems; Subclasses: containing "Relay / Switch" → Electromagnetic_Relay; containing "Bearing / Suspension" → Active_Magnetic_Bearing; otherwise Linear_Mag_Actuator;

[0119] "Medical / Robotics / Navigation / Thermotherapy" → Medical_and_Biomedical_Applications; Subclasses: Magnetically_Guided_MicroRobot / Magnetic_Hyperthermia / Magnetic_Drug_Targeting (select one based on semantics);

[0120] The target_type / target_mag empirical mapping is as follows: sensors are mostly metal_ferromagnetic; MRAM is mostly semiconductor_ferromagnetic; medical magnetic navigation is mostly metal_ferromagnetic; if it cannot be determined, retain the model output.

[0121] I / O example

[0122] Enter (short text): "Magnetic field sensor"

[0123] Output:

[0124] {"application_category":"Sensing_and_Detection","application_subcat":"Magnetostrictive_Stress_Sensor","target_type":"metal","target_mag":"ferromagnetic"}

[0125] Input (specification): "Tunnel junction for next-generation MRAM, requiring low write current"

[0126] Output:

[0127] {"application_category":"Electronics_and_Information_Storage","application_subcat":"MRAM_STT","target_type":"semiconductor","target_mag":"ferromagnetic"}

[0128] In this embodiment, the Few-shot prompt is designed as follows:

[0129] Input: "Design a high-sensitivity magnetic field sensor"

[0130] Output: {

[0131] "application_category": "Sensing_and_Detection",

[0132] "application_subcat": "Magnetostrictive_Stress_Sensor",

[0133] "target_type": "metal",

[0134] "target_mag": "ferromagnetic

[0135] }

[0136] S23: Based on the XGBoost prediction engine, receive the structured information extraction results and perform triple feature encoding to predict the material type and magnetic characteristics;

[0137] Feature engineering:

[0138] # Triple Feature Encoding

[0139] X['category_encoded'] = le_category.transform([category])

[0140] X['subcat_encoded'] = le_subcat.transform([subcat])

[0141] X['combo_encoded'] = le_combo.transform([f"{category}_{subcat}"])

[0142] Model architecture: XGBoost ensemble of 200 decision trees, max_depth=6, learning_rate=0.1

[0143] Multi-class output: 6 target combinations (metal / semiconductor / insulator × ferromagnetic / antiferromagnetic)

[0144] Model optimization: SMOTE oversampling and class weight balancing are used to address the data imbalance problem.

[0145] S24: The RAG-based intelligent recommendation generation subsystem performs precise matching queries based on the predicted material type and magnetic characteristics, uses a dynamic query algorithm to filter materials from the database, and generates a recommendation report that includes a description of the prediction results, an explanation of the technical principles, and relevant patent cases to support it.

[0146] Dynamic query algorithm:

[0147] SELECT * FROM application

[0148] WHERE target_type = ? AND target_mag = ?

[0149] ORDER BY confidence DESC LIMIT 5

[0150] Recommendation report generation:

[0151] First layer: Explanation of prediction results

[0152] Second layer: Explanation of technical principles

[0153] Third layer: Supporting patent cases

[0154] Quality control for generation: Limiting the length of patent abstracts to ensure that recommended text is concise and effective.

[0155] S3: Utilize the high availability system architecture module to implement multi-key polling technology using an API polling mechanism, thereby improving the stability and concurrency of large model calls; construct a multi-layered fault-tolerant processing system, and implement database query optimization (patent deduplication, pagination query, index optimization) and asynchronous front-end loading through a performance optimization engine to reduce user waiting time;

[0156] The multi-layered fault-tolerant processing subsystem in S3 includes full-process anomaly capture for input validation anomalies, model prediction anomalies, and database query anomalies, and is equipped with intelligent degradation strategies, as detailed below:

[0157] Unified Degradation Rules: Degradation is triggered if any step times out 3 times (1 / 2 / 4s exponential backoff); the reason for degradation / time / number of retries is recorded throughout the entire process for easy auditing and recovery.

[0158] Input validation error: Switch rule engine (domain keyword + co-occurrence mode) for judgment; if it cannot be judged, proceed with "pass with low confidence" (is_valid=true, confidence=0.6), and the front end prompts "semantic insufficiency, conservative judgment has been adopted".

[0159] Model prediction anomalies: Prioritize using the "most recent successful prediction cache of the same (cat, sub)" → fall back to the "most frequent combination in historical statistics" (e.g., metal_ferromagnetic) → then fall back to the "rule mapping table" (e.g., MRAM → semiconductor_ferromagnetic). Simultaneously label with a downgrade flag and confidence decay (-0.1 to -0.2).

[0160] Database query error: First read the read-only cache (the most recent 24-hour hit set / Top 20 materials) → Read the local read-only snapshot file (containing only mpid, formula, type, and mag) → If it still fails, only return "prediction result + patent basis" without the material list, and prompt the user to try again later.

[0161] Patent search error (if necessary): Change to "partial match search" (by target_type or target_mag only) or provide "general scenario description" as a fallback reason for recommendation.

[0162] User-visible strategy: On the results page, display a non-interrupted message indicating "Degradation strategy has been enabled. This will not affect browsing. We recommend refreshing later." Retain the refresh option.

[0163] S4: Utilize the intelligent user interaction module to provide a responsive web interface that supports user input of usage requirements, determines usage relevance, displays attribute combination prediction results and intelligent recommendation results, and provides a visual representation of the results.

[0164] Further includes:

[0165] Input requirements: The system adopts a responsive web interface design to support access from multiple devices and provides preset examples to guide and improve input quality;

[0166] Application relevance determination: AI-driven semantic analysis replaces simple keyword matching, providing user-friendly rejection prompts and application scenario descriptions;

[0167] Property combination prediction: High-precision material property prediction is achieved through an end-to-end prediction process that combines the XGBoost model with a three-feature fusion algorithm.

[0168] Intelligent recommendation generation: RAG technology is used to generate professional recommendation reasons, and the patent database is queried in real time to ensure the timeliness and accuracy of the recommendation basis;

[0169] Results visualization: A three-tiered display architecture is constructed, consisting of basic material information cards, intelligent recommendation analysis, and a material list, and it supports external links to the Materials Project details page to provide in-depth material information.

[0170] This invention achieves intelligent and precise recommendation services for two-dimensional magnetic materials through core technological innovations such as a multi-AI system collaborative architecture, the first application of RAG technology in material recommendation, a three-feature fusion XGBoost prediction model, a semantic understanding-based intelligent filtering mechanism, and real-time patent data-driven recommendation.

[0171] In one embodiment of the present invention, a user enters the requirement "magnetic tunnel junction material suitable for next-generation MRAM" on the platform homepage. The system first identifies the requirement as a two-dimensional magnetic material application scenario, extracts the keywords "MRAM" and "magnetic tunnel junction", and determines the required material property combination as "ferromagnetic semiconductor" through a semantic tag mapping model. The system then calls the built local database to filter out two-dimensional materials that meet the combined characteristics. The platform further displays the reasons for the recommendation and the corresponding 5 patent references, and lists 12 recommended materials, including the material chemical formula, MPID, and electrical and magnetic classifications. Users can click on the material name to be automatically redirected to the Materials Project website to view the material's detailed information.

[0172] Figure 2 Example text for users to input application scenarios of magnetic materials; Figure 3-5 The system recommends a results page, including the reasons for the recommendation, patent references, and material information; Figure 6 After a user clicks on a recommended material, they will be redirected to the Materials Project to view its detailed attributes. Figure 7-9 This provides an example of a feedback interface and suggested inputs for unrelated purposes.

[0173] In summary, this invention proposes an "application-driven" material reverse recommendation method: starting from the application requirements input by the user, it automatically predicts the matching material attribute combinations and constructs a complete reverse recommendation closed loop;

[0174] This invention integrates multi-source heterogeneous data: for the first time, it unifies and integrates structured two-dimensional material property data with unstructured application information in patent texts to construct a ternary correlation model of magnetic-electrical-application.

[0175] This invention introduces an interpretable recommendation mechanism: while outputting recommendation results, the system provides specific reasons for the recommendations and related patent literature references, thereby enhancing the credibility and traceability of the recommendations;

[0176] This invention constructs an online interactive platform that integrates usage identification, model prediction, and recommendation display functions. It has a good human-computer interaction experience, is easy to promote and apply, and has the potential for industrialization and commercialization.

[0177] Example 2, a reverse recommendation system driven by the application of two-dimensional magnetic materials, comprising:

[0178] The multi-source data fusion construction module includes a patent data intelligent annotation sub-module and a material property automatic calculation and classification sub-module, which are used to obtain patent data and property data corresponding to two-dimensional magnetic materials, construct a dual-table structure database, and obtain a multi-source fused database through an ETL data flow integration processing mechanism;

[0179] The multi-AI collaborative intelligent reasoning module includes an intelligent verification subsystem, a structured information extraction subsystem, an XGBoost prediction engine, and a RAG intelligent recommendation generation subsystem. It is used to receive user input regarding usage requirements, perform intelligent reasoning on them, and generate recommendation results.

[0180] The high availability system architecture module adopts an API polling mechanism to implement multi-key polling technology, builds a multi-layered fault-tolerant processing system, and achieves database query optimization and asynchronous front-end loading through a performance optimization engine.

[0181] The intelligent user interaction module provides a responsive web interface that supports user input of usage requirements, determines the relevance of usage, displays attribute combination prediction results and intelligent recommendation results, and provides a visual representation of the results.

[0182] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of the invention.

Claims

1. A method for reverse recommendation driven by the application of two-dimensional magnetic materials, characterized in that, Includes the following steps: S1: Use the multi-source data fusion construction module to obtain the patent data and attribute data corresponding to the two-dimensional magnetic materials, construct a dual-table database, and obtain the multi-source fused database through the ETL data stream integration processing mechanism; S2: Utilize a multi-AI collaborative intelligent reasoning module to receive user input regarding usage requirements, perform intelligent reasoning on them, and generate recommendation results; S2 includes the following steps: S21: Based on the intelligent verification subsystem, the relevantness of the received application requirements is judged, and intelligent prompts or rejection processing is performed for inputs with confidence levels below the threshold. S22: Extract relevant dimensional information from the verified usage requirements based on the structured information extraction subsystem; The structured information extraction subsystem in S22 receives valid input confirmed by the intelligent verification subsystem, extracts application category, application subcategory, target material type and target magnetic type dimension information, adds semantic reasoning prompts for inputs of less than 50 characters, and provides default reasoning logic based on keywords. S23: Based on the XGBoost prediction engine, receive the structured information extraction results and perform triple feature encoding to predict the material type and magnetic characteristics; S24: The RAG-based intelligent recommendation generation subsystem performs precise matching queries based on the predicted material type and magnetic characteristics, uses a dynamic query algorithm to filter materials from the database, and generates a recommendation report that includes a description of the prediction results, an explanation of the technical principles, and relevant patent cases to support it. S3: Utilizes a high-availability system architecture module to implement multi-key polling technology using an API polling mechanism, constructs a multi-layered fault-tolerant processing system, and achieves database query optimization and asynchronous front-end loading through a performance optimization engine; S4: Utilize the intelligent user interaction module to provide a responsive web interface that supports user input of usage requirements, determines usage relevance, displays attribute combination prediction results and intelligent recommendation results, and provides a visual representation of the results.

2. The method for driving reverse recommendation of two-dimensional magnetic material applications according to claim 1, characterized in that, S1 includes the following steps: S11: The intelligent annotation submodule based on patent data retrieves corresponding patent information from Google Patents Public Datasets by using SQL technology to query based on keywords related to two-dimensional magnetic materials. It then uses the Tongyi Qianwen big model and Few-shot prompting engineering technology to generate intelligent tags for the patent abstract. S12: The automatic calculation and classification submodule based on material properties obtains the magnetic information of two-dimensional materials from the Materials Project and the electrical information of two-dimensional materials from the 2DMatPedia database. S13: Through the material property calculation engine, the material type is automatically determined to be metal, semiconductor or insulator based on the band gap value, and ferromagnetism or antiferromagnetism is automatically identified through the magnetic sorting parameter. S14: Construct a dual-table structure of materials_magnetic and application tables, and establish an MPID standardized index; S15: Data cleaning is performed based on the ETL data stream integration processing mechanism, and the association mapping between patent data and material data is realized through MySQL.

3. The method for driving reverse recommendation based on the application of two-dimensional magnetic materials according to claim 2, characterized in that, In step S13, the material type is automatically determined as a metal, semiconductor, or insulator based on the band gap value. Specifically: A band gap value of 0 indicates that the material type is metal. A band gap value of 0-3 eV indicates that the material type is a semiconductor. If the band gap value is greater than 3eV, the material type is determined to be an insulator.

4. The method for driving reverse recommendation of two-dimensional magnetic material applications according to claim 1, characterized in that, The multi-level fault-tolerant processing system in S3 includes full-process anomaly capture for input validation anomalies, model prediction anomalies, and database query anomalies, and is equipped with intelligent degradation strategies.

5. A system for a reverse recommendation method driven by the application of two-dimensional magnetic materials as described in any one of claims 1-4, characterized in that, include: The multi-source data fusion construction module includes a patent data intelligent annotation sub-module and a material property automatic calculation and classification sub-module, which are used to obtain patent data and property data corresponding to two-dimensional magnetic materials, construct a dual-table structure database, and obtain a multi-source fused database through an ETL data flow integration processing mechanism; The multi-AI collaborative intelligent reasoning module includes an intelligent verification subsystem, a structured information extraction subsystem, an XGBoost prediction engine, and a RAG intelligent recommendation generation subsystem. It is used to receive user input regarding usage requirements, perform intelligent reasoning on them, and generate recommendation results. The high availability system architecture module adopts an API polling mechanism to implement multi-key polling technology, builds a multi-layered fault-tolerant processing system, and achieves database query optimization and asynchronous front-end loading through a performance optimization engine. The intelligent user interaction module provides a responsive web interface that supports user input of usage requirements, determines the relevance of usage, displays attribute combination prediction results and intelligent recommendation results, and provides a visual representation of the results.

Images