A method and system for generating a scientific and technological innovation big data model based on a big model

Abstract

The present invention relates to a method and system for generating a scientific and technological innovation big data model based on a large model, and belongs to the field of artificial intelligence and big data processing. The method comprises: extracting and collating raw data from various scientific research data sources, and constructing a graph structure containing domain knowledge after denoising and standardization. At the same time, the system integrates a long-term memory module, continuously updates and stores historical data, and supports dynamically adjusted learning algorithms to adapt to different data inputs. Based on the data in long-term memory, a generative model is used to generate innovative ideas, and the generation process is optimized in combination with reinforcement learning technology. Finally, sentiment analysis and topic modeling are performed by collecting user feedback information to further optimize the system parameters and algorithms.

Description

Technical Field

[0001] The present invention relates to the field of big data processing and artificial intelligence technology, and in particular to a method and system for generating a scientific and technological innovation big data model based on a big model. Background Art

[0002] Traditional scientific research data analysis methods often rely on manual collation and analysis of large amounts of scientific research literature, patent documents, conference reports and other data. This method is not only time-consuming and labor-intensive, but also difficult to systematically capture interdisciplinary knowledge connections, limiting the ability of researchers to discover innovative ideas. In addition, with the rapid development of science and technology, scientific research data has exploded, and traditional literature reviews and expert experience can no longer effectively handle such a large amount of data. At present, although there are some attempts to use computers to assist scientific research innovation, most of these methods have the following problems:

[0003] (1) The data source is single, making it difficult to fully cover interdisciplinary scientific research information;

[0004] (2) Limited data processing capabilities, unable to effectively clean, standardize, and integrate diverse data from different sources;

[0005] (3) The lack of an effective mechanism to build and maintain structured domain knowledge graphs limits knowledge discovery and innovation generation;

[0006] (4) The long-term memory mechanism is imperfect, which makes the model unable to effectively accumulate and utilize historical knowledge;

[0007] (5) Insufficient adaptive learning capabilities make it difficult for the model to self-adjust and optimize based on new data and new knowledge. Summary of the invention

[0008] The present invention provides a method and system for generating a scientific and technological innovation big data model based on a large model, so as to solve the problem of how to build an efficient knowledge graph by integrating various scientific research data sources, and generate innovative ideas and continuously optimize system performance based on long-term memory and generation models.

[0009] In order to solve the above problems, the present invention provides a method for generating a scientific and technological innovation big data model based on a big model, comprising:

[0010] Extract and organize raw data from various scientific research data sources, perform denoising and standardization, and build a graph structure containing domain knowledge;

[0011] Integrate long-term memory modules in the system to continuously update and store historical data, support dynamically adjusted learning algorithms, and adapt to different data inputs;

[0012] Generate innovative ideas using generative models based on data in long-term memory, optimize the generation process with reinforcement learning techniques, and adjust the generation strategy through feedback;

[0013] Collect user feedback information, conduct sentiment analysis and topic modeling, and optimize the system parameters and algorithms.

[0014] Furthermore, before the step of extracting and arranging the original data, it also includes:

[0015] Obtain key technical terms and hot topics in the field related to the scientific research data source.

[0016] Furthermore, the step of constructing the graph structure of domain knowledge specifically includes:

[0017] Build an index of technical terms and monitor index build time and index size.

[0018] Furthermore, the integration step of the long-term memory module specifically includes:

[0019] Historical data is categorized and stored, and a dynamically adjusted learning algorithm is set up.

[0020] Furthermore, the steps of generating innovative ideas by the generative model specifically include:

[0021] The data in the long-term memory is input into the generative model to generate innovative ideas to be optimized.

[0022] Furthermore, the step of optimizing the generation process by combining reinforcement learning technology specifically includes:

[0023] Adjust generation strategies based on user feedback to improve the quality of innovative ideas.

[0024] Furthermore, the step of collecting user feedback information specifically includes:

[0025] User feedback on the generated ideas was collected through questionnaires and online reviews.

[0026] Furthermore, the steps of sentiment analysis and topic modeling specifically include:

[0027] Perform sentiment analysis on user reviews to identify positive and negative feedback and summarize opinion points.

[0028] Furthermore, the step of optimizing system parameters and algorithms specifically includes:

[0029] The parameters of the long-term memory module and the generation model are adjusted according to user feedback and sentiment analysis results.

[0030] Furthermore, the optimization steps of the parameters and algorithms specifically include:

[0031] Generate comprehensive evaluation reports regularly to analyze system operation status and performance indicators to guide subsequent system improvements.

[0032] The main beneficial effects of the present invention are as follows:

[0033] (1) Comprehensive coverage of scientific research data: The present invention automatically obtains scientific research data from multiple sources, including academic papers, patent documents, conference reports, open source projects, industry reports, etc., ensuring the comprehensiveness and diversity of data, thereby providing rich and multi-faceted support for scientific research innovation.

[0034] (2) Efficient data preprocessing: By performing preprocessing operations such as cleaning, deduplication and standardization on the collected data, the data quality is improved, noise interference is reduced, and the accuracy of subsequent processing is ensured.

[0035] (3) Constructing high-quality knowledge graphs: Natural language processing techniques (such as word embedding, entity recognition, and relationship extraction) are used to extract key information from text and construct structured domain knowledge graphs, which help the model better understand the complex relationships and background knowledge in the scientific research field.

[0036] (4) Support for long-term memory and adaptive learning: By maintaining long-term memory and supporting adaptive learning mechanisms, the present invention can continuously adjust and optimize the content of long-term memory, ensuring that the model can continue to improve with the development of time and new knowledge, providing a solid foundation for generating innovative ideas.

[0037] (5) Generation of innovative ideas: We use long-term memory to generate innovative ideas and optimize the generation process through technologies such as reinforcement learning, thereby improving the quality and practicality of generated ideas and promoting the efficiency and effectiveness of scientific research innovation.

[0038] (6) Efficient query and access services: It provides a flexible and easy-to-use API interface that supports multiple query methods, enabling other modules to efficiently query and access information in the knowledge graph, thereby speeding up the acquisition and utilization of scientific research information.

[0039] (7) Continuous updating and optimization: The knowledge graph is updated regularly to reflect the latest research results and development trends, ensuring the timeliness and accuracy of the knowledge graph. At the same time, query performance and maintenance strategies are continuously optimized to ensure the stable operation of the system.

[0040] (8) Security and privacy protection: Security authentication and permission management functions are implemented to ensure that only authorized users or modules can access specific data, thus protecting sensitive information in the knowledge graph. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1It is a flow chart of a method for generating a scientific and technological innovation big data model based on a big model provided in an embodiment of the present invention.

[0042] Figure 2 It is a flow chart of the data integration and knowledge graph construction module provided in an embodiment of the present invention.

[0043] Figure 3 It is a flow chart of the long-term memory and adaptive learning module provided in an embodiment of the present invention.

[0044] Figure 4 It is a flow chart of the innovation generation and optimization module provided in an embodiment of the present invention.

[0045] Figure 5 It is a flowchart of the feedback and iterative improvement module provided in an embodiment of the present invention.

[0046] Figure 6 This is a system flow chart of the method for generating a scientific and technological innovation big data model based on a big model of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0048] In the following, only some exemplary embodiments are briefly described. As those skilled in the art will appreciate, the described embodiments may be modified in various ways without departing from the spirit or scope of the present application. Therefore, the drawings and descriptions are considered to be exemplary and non-restrictive in nature.

[0049] Reference Figure 1 , is a flowchart of a method for generating a scientific and technological innovation big data model based on a big model according to an embodiment of the present invention, and the flowchart may at least include steps S100-S400:

[0050] S100, extract and organize raw data from various scientific research data sources, perform denoising and standardization, and build a graph structure containing domain knowledge;

[0051] S200, integrates long-term memory modules into the system to continuously update and store historical data, and supports dynamically adjusted learning algorithms to adapt to different data inputs;

[0052] S300, based on the data in long-term memory, use the generative model to generate innovative ideas, and combine reinforcement learning technology to optimize the generation process and adjust the generation strategy through feedback;

[0053] S400, collect user feedback information, conduct sentiment analysis and topic modeling, so as to continuously optimize the system parameters and algorithms and improve the overall performance.

[0054] like Figure 2 As shown, step S100 includes steps S110-S160:

[0055] S110, automatically acquire scientific research data from multiple sources, including but not limited to academic papers, patent documents, conference reports, open source projects, industry reports, etc. These data sources cover a wide range of disciplines, ensuring the comprehensiveness and diversity of the data. Specifically:

[0056] Get the latest research papers from major academic databases and journals such as IEEEXplore, Springer, Elsevier, PubMed, etc. These papers usually contain the latest scientific discoveries and technological advances.

[0057] Patent documents are collected through patent databases such as the United States Patent and Trademark Office (USPTO), the European Patent Office (EPO), etc., which record the specific details of technological innovations and inventions.

[0058] Obtain reports and abstracts of important international and domestic academic conferences, which provide the latest research directions and technological trends in the field.

[0059] Crawl code, documents, and discussions of open source projects from code hosting platforms such as GitHub and GitLab. These projects often reflect best practices and technical challenges in actual development.

[0060] Understand industry dynamics and technological development trends through industry analysis reports, market research reports, etc. released by authoritative organizations.

[0061] S120: Perform pre-processing operations such as cleaning, deduplication, and standardization on the collected data to improve data quality and reduce noise interference. Specifically:

[0062] Identify and correct or delete erroneous, incomplete, inaccurate or irrelevant data entries. Data indicators of interest include but are not limited to scientific research papers, patent documents, conference reports, open source projects and industry reports. Therefore, in the process of data cleaning, focus on checking whether the data in these sources are complete and contain key information (such as author, publication date, keywords, etc.), and remove missing values ​​or erroneous information.

[0063] The same information may be obtained repeatedly during the data collection process, especially when dealing with a large number of sources. Therefore, deduplication needs to be performed to ensure the uniqueness of each data entry. This step is critical to keeping the dataset refined and efficient. For data sources such as patent documents and academic papers that are prone to duplication, duplicate entries are identified and merged based on key fields such as title, abstract, and author list.

[0064] Standardization refers to converting data into a unified format to facilitate subsequent data integration and analysis. For example, different date formats can be converted into the ISO standard format (YYYYMMDD), or natural language processing operations such as word segmentation and stop word removal can be performed on text data. When processing scientific research data, standardization also includes unified definitions of terms and concepts to avoid misunderstandings caused by terminology differences.

[0065] During the preprocessing process, special attention will be paid to the core data indicators related to the "Long-term Memory and Innovation Generation" module. This means that during the data cleaning, deduplication and standardization process, information that helps the model understand and generate truly novel and creative ideas is retained. For example, ensure that data that can reflect scientific and technological innovation trends, research hotspots and potential technological breakthroughs are retained, while eliminating irrelevant or redundant information.

[0066] The data quality is continuously monitored throughout the preprocessing phase to ensure that all operations meet the predetermined standards. For example, thresholds are set to evaluate indicators such as data integrity and consistency to ensure that the preprocessed data can meet the needs of subsequent processing.

[0067] S130. Extract key information and concepts from texts through natural language processing (NLP) techniques such as word embedding, entity recognition, and relationship extraction. Specifically:

[0068] Word embedding is a technique that maps words to multidimensional vector spaces. These vectors can capture the semantic relationship between words. When processing data such as scientific research papers, patent documents, conference reports, etc. that invention patents focus on, word embedding can help understand the professional terms and concepts in the text. For example, by training a word embedding model, a vector space can be obtained in which similar terms are mapped to similar positions, which helps with subsequent relationship extraction and entity recognition.

[0069] Entity recognition (NER, Named Entity Recognition) refers to the automatic identification of named entities in text, such as names of people, organizations, places, etc. In application scenarios, entity recognition can help identify key figures, research institutions, technical terms, etc. in scientific research literature. These entities will serve as nodes in the knowledge graph and become the basic units for building knowledge networks.

[0070] Relation extraction is the process of extracting the relationships between entities from text, which is crucial for building a knowledge graph. For example, you can identify the relationship between an inventor and his organization, or the relationship between a technology and its application area. Relation extraction helps understand the interactions between different entities and encodes them as edges in the knowledge graph.

[0071] During the feature extraction process, special attention will be paid to the core data indicators related to the "Long-term Memory and Innovation Generation" module. This means that in the process of word embedding, entity recognition, and relationship extraction, emphasis will be placed on extracting information that helps the model understand and generate truly novel and creative ideas. For example, entity recognition technology is used to identify key figures and technical terms that represent technological innovation trends, and relationship extraction technology is used to understand the connections between them.

[0072] In addition to basic word embedding and entity recognition, more advanced natural language processing techniques are also used to parse the deeper meaning of the text. This includes but is not limited to dependency syntactic analysis, semantic role labeling, etc. These technologies can help better understand the logical relationships and event sequences in the text, thereby providing support for building more complex and sophisticated knowledge graphs.

[0073] S140, using the extracted information to construct a domain knowledge graph, including the relationship network between entities, attribute information and their interactions, to form a structured knowledge representation. Specifically:

[0074] First, based on the results of the feature extraction phase, define the entity categories and relationship types in the graph. For example, when processing data related to "long-term memory and innovation generation", entities may include inventors, research institutions, technical concepts, patent documents, etc., while relationships may include inventors belonging to a certain institution, technical concepts being applied to a specific field, and one patent citing another patent.

[0075] Based on the results of entity recognition, nodes in the knowledge graph are created. Each node corresponds to a specific entity, such as an inventor or a technology. The node contains not only basic information about the entity, but may also contain more details, such as the inventor's work experience, description of the technical concept, etc.

[0076] Based on the results of relationship extraction, connecting edges are added between entities, that is, the relationships between entities are defined. These relationship edges reflect various associations between entities, such as affiliation, citation, cooperation, etc. For example, if relationship extraction finds that a patent cites another patent, a citation edge will be established between the corresponding patent nodes.

[0077] Each node and edge can carry rich attribute information, which comes from the context and metadata in the original text. For example, for a technical concept node, its attributes may include technical description, time of first appearance, main application areas, etc. This attribute information helps to enhance the expressiveness of the graph.

[0078] In order to ensure the quality of the knowledge graph, a series of optimization operations need to be performed on the graph. This includes but is not limited to eliminating redundant information, solving synonym problems, filling in missing attributes, etc. For example, through synonym recognition technology, different term nodes with the same or similar meanings can be merged, thereby reducing the complexity of the graph and improving its consistency.

[0079] As new data is continuously added, the knowledge graph will continue to be updated and expanded. New entities, relationships, and attribute information will be incorporated into the graph, which will help the model understand a wider range of background knowledge and generate innovative ideas based on this.

[0080] In the process of building the knowledge graph, special attention will be paid to the core data indicators related to the "Long-term Memory and Innovation Generation" module. This means that based on the information obtained in the feature extraction phase, entities and relationships that help the model understand and generate truly novel and creative ideas will be constructed. For example, special attention may be paid to entity nodes that represent technological innovation trends and the relationships between them may be emphasized so that the model can better understand and use this information to generate innovative solutions.

[0081] S150. Update the knowledge graph regularly to reflect the latest research results and development trends, and ensure the timeliness and accuracy of the knowledge graph. Specifically:

[0082] Continuously obtain the latest information from multiple sources such as patent databases, academic journals, conference proceedings, industry reports, etc. These data will be pre-processed, including cleaning, standardization and formatting, to ensure data quality. In particular, for this invention, special attention will be paid to data indicators that can promote long-term memory and innovation generation, such as information related to technological innovation trends, emerging technological concepts, important inventors and their contributions.

[0083] Feature extraction is performed on the newly added data, including entity recognition, relationship extraction, and attribute extraction. These feature extraction processes will use natural language processing technology and machine learning algorithms to ensure that key information in the latest data can be accurately captured. For example, entity recognition will help discover new inventors, technical concepts, etc., while relationship extraction can help understand the connection between these new entities and other known entities.

[0084] Integrate the new entities, relationships, and attribute information obtained from incremental feature extraction into the existing knowledge graph. This step includes adding new nodes, updating the attribute information of existing nodes, and creating or modifying the connection edges between nodes. For example, if new data reveals the application of a technology in a certain field, a new attribute will be added to the graph to indicate its application in this field.

[0085] During the update process, you may encounter inconsistent information, such as synonym problems of entities, differences in attribute information, etc. To this end, you need to implement conflict detection and resolution strategies to ensure the consistency and accuracy of the knowledge graph. This may include using semantic similarity calculations to identify synonyms and adopting expert review or majority voting to decide the final entity representation.

[0086] The updated knowledge graph is quality-controlled through a combination of manual review and automatic verification. This includes checking the correctness of entities and relationships, the completeness of attribute information, and the overall coherence of the graph. In addition, graph optimization is performed to reduce redundant information and improve query efficiency.

[0087] Develop a periodic maintenance plan to ensure that the knowledge graph can regularly receive the latest data and be updated accordingly. The maintenance cycle can be determined based on the update frequency and importance of the data source, such as a comprehensive maintenance every quarter or every year.

[0088] Throughout the maintenance process, special attention will be paid to data indicators that can promote long-term memory and innovation generation. For example, we will continue to track technological innovation trends, the development of emerging technological concepts, the activities of important inventors, and their influence in different fields. This information will be integrated into the knowledge graph to support the model to better understand existing knowledge and generate truly novel and creative ideas based on this.

[0089] S160, provide interfaces and services to enable other modules to efficiently query and access information in the knowledge graph. Specifically:

[0090] Design and implement a flexible and easy-to-use API interface that supports multiple query methods, including but not limited to precise query, fuzzy query, and complex query (such as based on SPARQL syntax). These query functions are designed to help other modules quickly locate and obtain the required knowledge graph information. For example, in order to support the needs of the present invention, it is necessary to be able to quickly retrieve entities, relationships, and their attributes related to long-term memory and innovation generation.

[0091] When other modules send query requests, the knowledge graph query module will first parse the request and extract the key elements of the query, such as the query target, query scope, query conditions, etc. For example, if the query is about the application field of a specific technology, the query request will contain relevant keywords such as the technology name and application field.

[0092] Based on the parsed query request, the system will perform the corresponding query operation. This usually involves the use of database indexes to increase query speed. For complex query requests, the system may also need to perform multiple sub-queries and synthesize the results. For example, in order to find out the development trend of a certain technology in different time periods, the system needs to query the relevant nodes of the technology and extract the attribute values ​​that change over time.

[0093] After the query is completed, the system will process the query results, such as deduplication, sorting, and filtering, and then return the processed results to the requester. For example, when querying research trends related to long-term memory and innovation generation, the system may return a series of research results arranged in chronological order, including relevant patents, academic papers, etc.

[0094] In order to ensure query efficiency, the system will regularly monitor and optimize query performance. This includes but is not limited to measures such as establishing and updating indexes, optimizing query algorithms, and caching commonly used query results. For example, for frequently queried technical concepts, the system can pre-calculate its associated entity network and cache its results for quick response.

[0095] In order to protect sensitive information in the knowledge graph, the query interface also needs to implement security authentication and permission management functions. Only authorized users or modules can access specific data. For example, for information containing commercial secrets or personal privacy, only users with corresponding permissions can access it.

[0096] like Figure 3 As shown, step S200 at least includes steps S210-S260:

[0097] S210, store and manage the structured data and knowledge graph obtained from the data integration and knowledge graph construction module as part of the long-term memory. Specifically:

[0098] When the data integration and knowledge graph construction module completes data processing and builds the knowledge graph, these structured data will be passed to the long-term memory management module. First, the module will perform integrity checks and format verification on the incoming data to ensure that all incoming data conforms to the preset standard format, such as ensuring that each entity has a unique identifier and that the relationship between entities has been correctly defined.

[0099] The verified data is then stored in a specially designed database, which is usually in the form of a graph database to facilitate efficient management and query of knowledge graphs. For example, considering the needs of invention patents, that is, to understand and generate novel ideas in a large model, with a special focus on entities related to innovation generation and the connections between them, these data will be stored and labeled first.

[0100] To improve query efficiency, the long-term memory management module automatically creates indexes for key entities and relationships in the knowledge graph. In addition, data optimization is performed regularly, such as merging duplicates, deleting entities that are no longer relevant, etc. The purpose of this is to ensure that the content of long-term memory is always up-to-date and easy to access. For example, for entities with high frequency of queries, such as key technologies, important people, or events, the system will create additional indexes to speed up the query process.

[0101] The long-term memory management module also needs to regularly check and maintain the stored data, including but not limited to data backup, recovery strategy, and data lifecycle management. For example, in order to ensure the security and persistence of data, the system will implement regular data backup and set up an automatic recovery mechanism to deal with unexpected situations.

[0102] Since the information in the knowledge graph changes as new data is added, the long-term memory management module records every modification and update of the data to form a version control system. This helps to track historical changes and also facilitates tracing back to previous states. For example, when an attribute of an entity changes, the system records the timestamp of the change, the person who modified it, and the specific content of the change.

[0103] To ensure the quality of long-term memory, the module implements a data cleaning process to remove inconsistent or erroneous data entries. For example, the system will clean or mark out outdated or corrected information to avoid affecting subsequent analysis and decision-making.

[0104] S220. Continuously adjust and optimize the content of long-term memory based on new data and knowledge to ensure that the model can continue to improve with the development of time and new knowledge. Specifically:

[0105] After the new data and knowledge are processed by the data integration and knowledge graph construction module, these structured data and updated knowledge graphs will be passed to the adaptive learning module. First, the module will pre-process the new data, including cleaning, standardization and formatting, to ensure that it is compatible with the existing long-term memory.

[0106] After new data is processed, the module starts an incremental learning process to gradually integrate the new knowledge into the existing long-term memory. Incremental learning allows the model to be trained only on new data without retraining the entire model, saving computing resources. For example, when receiving new data about emerging technology trends, the model will focus on learning the relevant entities of these trends and their relationships with entities in the existing knowledge graph.

[0107] The module uses logical reasoning and association analysis algorithms to evaluate the relationship between new data and existing long-term memory. This step helps identify potential patterns, trends, and anomalies, thereby providing the model with richer background knowledge. For example, if the new data contains a breakthrough discovery in a certain field, the model will try to infer the possible impact of this discovery and update the relationship between related entities.

[0108] After analyzing the new data in relation to existing knowledge, the model updates its internal representation based on these analysis results. This includes, but is not limited to, updating nodes and edges in the knowledge graph, adjusting weights, and adding new nodes or edges. For example, if the model detects that technology in a certain field is developing rapidly, it may increase the weight of that field and adjust the importance of related entities accordingly.

[0109] To ensure continuous improvement of the model, the adaptive learning module regularly evaluates the model's performance when processing new data and makes necessary adjustments based on the evaluation results. This includes monitoring the model's accuracy, recall and other indicators when processing specific types of data. For example, for data involving technological innovation, the model will pay special attention to its ability to predict future development trends and adjust the learning strategy accordingly.

[0110] Adaptive learning is an ongoing process. As more data is accumulated and knowledge is updated, the model will continue to learn and adjust itself. This means that the model can continue to improve over time and better understand and generate novel ideas. For example, as more and more data on technological innovation is integrated, the model can not only capture the latest technological trends, but also generate forward-looking innovation suggestions based on this.

[0111] S230, supports personalized learning paths for different users or application fields, and can adjust learning strategies and content according to specific needs. Specifically:

[0112] The personalized learning module first needs to collect user preference information and data in specific application areas. This includes but is not limited to user interests, past interaction records, expertise in specific fields, and technical documents. For example, for a user who focuses on the field of biotechnology, the module will give priority to patent literature, research papers, conference reports, etc. in this field as input data.

[0113] Based on the information collected, the module builds a preference model of the user or application domain. This usually involves using machine learning algorithms to identify key features and preference indicators. For example, for a user focused on biotechnology, the module may identify key words that the user frequently reads (such as gene editing, protein engineering), the technology trends that the user pays attention to (such as the development of CRISPR technology), and the projects or topics discussed by the user.

[0114] Based on the preference model, the module will filter out the data and knowledge that are most relevant to the user or application field from the big data pool. These data may include the latest scientific research results, industry reports, technical white papers, etc. For example, for users in the above biotechnology field, the module may recommend articles or papers on the recent progress of gene editing technology.

[0115] The module will set a personalized learning path based on the user's preferences and goals. This includes determining which knowledge areas need to be focused on and which skills need to be improved. For example, if the user wants to gain a deeper understanding of the application of gene editing technology, the module will plan a learning path that guides the user to gradually master the basics and then delve into specific cases and technical details.

[0116] As user interactions and feedback accumulate, the module will continuously adjust its learning strategy to better meet user needs. This includes updating preference models, adjusting the weight of recommended content, and optimizing learning paths. For example, if a user frequently views articles about CRISPRCas9 technology, the module will automatically increase the frequency of recommendations for materials related to the technology.

[0117] The personalized learning module continuously monitors the user's learning progress and satisfaction and adjusts the learning strategy based on this feedback. This includes regularly collecting user feedback, evaluating the learning effect and making corresponding adjustments. For example, the effectiveness of the learning path can be evaluated by analyzing the user's dwell time on a specific topic, click behavior and completion of the course.

[0118] In addition to providing personalized learning content, the module will also use the accumulated knowledge base to stimulate users' innovative thinking. This can be achieved by asking open questions and providing cross-disciplinary knowledge links. For example, users can be provided with information about the intersection of gene editing technology and other fields to inspire them to think about new application scenarios or solutions.

[0119] S240. Make inferences and logical inferences based on information in long-term memory to help generate new insights and hypotheses. Specifically:

[0120] The knowledge reasoning module first needs to build a comprehensive knowledge graph, which contains data from different sources, such as patent documents, research papers, conference reports, etc. These data form the basis of long-term memory, including the core data indicators collected by the present invention, such as key technical terms, hot topics in the field, and emerging technology trends.

[0121] Based on the constructed knowledge graph, the module will identify entities and extract the relationships between entities. For example, in the field of biotechnology, entities can be gene editing technology, CRISPRCas9 system, gene therapy, etc., while relationships can be the application of technology, the relationship between technologies, etc.

[0122] The module uses entities and relationships in the graph to perform logical reasoning. This includes direct reasoning (logical deduction based on known facts) and indirect reasoning (inference based on the connection between multiple facts). For example, if the graph contains the development history of gene editing technology and the latest progress of the CRISPRCas9 system, the module can predict the future direction of technology development or potential application scenarios through logical reasoning.

[0123] The module analyzes the information in the knowledge graph based on the current context to generate more specific hypotheses or insights. For example, if the current context is that a biotech company is looking for new gene editing technologies to treat genetic diseases, the module can combine the existing knowledge graph to recommend specific technology paths or potential partners.

[0124] In order to generate more novel ideas, the module will also explore cross-disciplinary connections. For example, combining gene editing technology with nanomaterial science to explore their potential applications in drug delivery systems. This cross-disciplinary knowledge integration helps discover new innovations.

[0125] Based on the above reasoning process, the module generates a series of hypotheses or new insights. These hypotheses can be about the possibility of new technologies, new uses of existing technologies, or ways to improve existing solutions. For example, the module may propose a hypothesis that a new type of gene therapy can be developed using the CRISPRCas9 system to treat rare genetic diseases.

[0126] Finally, the module will conduct a preliminary evaluation and verification of the generated hypothesis. This includes checking whether the hypothesis is consistent with existing scientific principles, whether there is sufficient evidence to support it, and whether it has practical application prospects. For example, for the proposed new gene therapy hypothesis, the module will check whether similar research results have been published and whether there are any ethical or technical obstacles that may be faced.

[0127] S250: Identify relevant knowledge and background information in specific situations to provide richer and more targeted support for innovation generation. Specifically:

[0128] First, the module needs to identify the current context. This may include the specific problem domain, target user group, technical requirements, or other relevant contextual conditions. For example, if considering the biotech field, the context may be "Developing a gene therapy for a rare genetic disease."

[0129] After clarifying the context, the module retrieves data related to the context from long-term memory. These data include but are not limited to patent documents, research papers, industry reports, etc., especially the core data indicators collected in this invention. These indicators cover key technical terms, hot topics in the field, emerging technology trends, etc.

[0130] The module then analyzes the input data to determine which information is most relevant to the current context. For example, if the context is about gene therapy research and development, data related to CRISPRCas9 technology, gene editing methods, genetic disease treatment cases, etc. will be prioritized.

[0131] The module extracts key background information from relevant data, including technology status, market trends, regulatory environment, etc. This background information helps to understand the situation more deeply and lays the foundation for subsequent innovation generation.

[0132] The module converts the extracted information into contextual features, which can be statistical indicators, technology maturity, application scope, etc. For example, if the context is the development of gene therapy, then the features may include the number of successful cases of CRISPRCas9 technology, the safety evaluation criteria of gene editing technology, etc.

[0133] The module will match the constructed situational features with the information in long-term memory to find the most relevant pieces of knowledge. For example, if the situational features indicate the need for an efficient and safe gene editing tool, the module will look for safety studies, clinical trial results, etc. related to CRISPRCas9.

[0134] Finally, the module feeds back the matched relevant knowledge and background information to the innovation generation process. This information not only provides rich details, but also ensures that the generated innovation concepts are highly consistent with the situation. For example, if the situation is innovation in the field of gene therapy, the module will provide successful cases of CRISPRCas9 in treating genetic diseases, as well as current challenges and future development possibilities.

[0135] S260. Regularly evaluate the effectiveness of long-term memory and learning strategies and make necessary adjustments based on the evaluation results. Specifically:

[0136] Before starting the assessment, the module needs to define the objectives and criteria of the assessment. These objectives are usually related to a specific situation or task, such as improving the quality of innovative proposals in a certain technical field or enhancing the understanding of emerging technological trends.

[0137] In order to conduct effective evaluation, the module first needs to integrate data from different sources. These data include but are not limited to the core data indicators collected in the present invention. These indicators cover key technical terms, hot topics in the field, emerging technology trends, etc.

[0138] Based on the input data and context, the module will select appropriate performance indicators to measure the learning effect. For example, when evaluating the model's understanding of the biotechnology field, it may focus on the following aspects:

[0139] Accuracy: Evaluate whether the information provided by the model is accurate.

[0140] Novelty: Evaluate whether the ideas generated by the model are novel and unique.

[0141] Practicality: Evaluate whether the recommendations made by the model are feasible and can be applied in real-world scenarios.

[0142] Relevance: Evaluate how relevant the content generated by the model is to the current context.

[0143] The module uses the selected performance indicators to evaluate the performance of the model. This may involve a variety of methods such as simulated scenario testing, expert review, user feedback, etc. For example, a series of cases in the field of biotechnology can be designed, and the model can provide solutions, and then the answers of the model can be scored by domain experts.

[0144] For each round of evaluation results, the module conducts data analysis to identify the strengths and weaknesses of the model. For example, if the model scores low on the novelty metric, it may mean that it has limitations in generating innovative ideas.

[0145] Based on the evaluation results, the module adjusts the long-term memory and learning strategies. For example, if the model is found to perform poorly in a certain area, it can be improved by introducing more training data in that area. In addition, algorithm parameters can be adjusted or different learning methods can be used to improve performance.

[0146] Learning effectiveness evaluation is an iterative process, and the module will repeat the above steps regularly to continuously optimize its performance. For example, you can set up a comprehensive evaluation every quarter and make corresponding adjustments based on the latest evaluation results.

[0147] like Figure 4 As shown, step S300 at least includes steps S310-S350:

[0148] S310: Use information in long-term memory to generate novel ideas and solutions through generative models (such as variational autoencoders, generative adversarial networks, etc.). Specifically:

[0149] S311. For each piece of input data , first perform feature extraction to obtain the feature vector:

[0150]

[0151] in, Represents the jth feature of the i-th data.

[0152] S312, in the preprocessing stage, all feature vectors are standardized to the same scale to eliminate the dimension effect. and are the mean and standard deviation of the jth feature in all data, respectively. For each feature , perform the following transformation:

[0153]

[0154] S313, an original generative model based on variational autoencoder (VAE) is adopted. The model consists of two parts: encoder and decoder, where the encoder maps the input data to the latent space, and the decoder generates new data points from the latent space.

[0155] Let z be the latent variable and θ be the encoder parameter. Then the goal of the encoder is to approximate the posterior probability . Assume that z has mean μ and variance The encoder outputs μ and log( ). Specifically, the output of the encoder can be expressed as:

[0156]

[0157] in, is the preprocessed feature vector, , and , are the weights and bias terms of the encoder.

[0158] The goal of the decoder is to recover the original data from the latent variable z. is the decoder parameter, then the output of the decoder is It can be expressed as:

[0159]

[0160] Among them, g is the activation function, and are the weights and biases of the decoder.

[0161] S314, in order to evaluate the innovation of generated content, an evaluation function S( , ), which measures the original data With generated data The difference between . Define S as:

[0162]

[0163] in, yes The j-th normalized feature of .

[0164] S315. In order to train the model, a loss function L including reconstruction error and KL divergence is defined:

[0165]

[0166] in, represents the KL divergence, p(z) is the prior distribution of the latent variable z, which is a standard normal distribution. Minimize L by gradient descent and update the parameters θ and θ of the encoder and decoder. .

[0167] After training is complete, randomly sample z and generate new data points through the decoder . This process can be expressed as:

[0168]

[0169] For each generated data point , use the innovation evaluation function S to evaluate, and select those data points with higher innovation as the final innovative ideas.

[0170] Generated data And its evaluation result S is fed back into long-term memory for future model training, thereby continuously optimizing model performance.

[0171] S320: Use reinforcement learning algorithms to guide the process of generating innovative ideas, and use reward mechanisms to encourage models to explore more valuable and innovative solutions. Specifically:

[0172] S321. Input data is derived from the content of the invention, including key technical terms, hot topics in the field, emerging technology trends, etc. These data form the basis for the generation of innovative ideas. Environmental status It is composed of the current data set and its features, namely ,in represents the i-th feature vector at time t. These feature vectors can be obtained from long-term memory or new features obtained through feature extraction.

[0173] action Indicates the action taken by the model in the current state, which can be to adjust model parameters, change the combination of data sets, etc. Specifically, the action can be defined as a vector in space ,in Represents the adjustment of the j-th model parameter or data processing method.

[0174] The policy π defines the probability distribution of actions to choose in a given state. In the context of reinforcement learning, the policy can be deterministic or stochastic. Here we use a stochastic policy, π( | ).

[0175] award Used to evaluate the quality of an action. For the generation of innovative ideas, the reward function It can be defined based on the innovation and practicality of the generated ideas. The following formula will be introduced to calculate the reward function.

[0176] S322, original algorithm formula:

[0177] ① First, define a multiple regression equation to predict the potential value of innovative ideas:

[0178]

[0179] in ( ) is in the state The predicted value of is a constant term, Is with characteristics The coefficient of correlation.

[0180] ②Use partial differential equations to describe the change of strategy π with time t:

[0181]

[0182] Where f is a nonlinear function that can be used according to the current state and actions to adjust strategy.

[0183] ③ We also use the concept of complex analysis to define the reward function The calculation formula is Function g(z) that depends on the complex variable z:

[0184]

[0185] Where C is a closed path and g(z) is a complex function related to innovation and practicality, which is defined as:

[0186]

[0187] Where I(z) represents the plural form of the innovation score, U(z) represents the plural form of the practicality, and is the weighting coefficient.

[0188] S323, reinforcement learning process: Initialize model parameters, strategy π and value function V( ). Set the initial state is the input data set.

[0189] Select actions based on the current strategy π .

[0190] Apply Action Update Status To the next state .

[0191] According to the status Calculating Rewards . This is calculated by solving the complex integral .

[0192] Use the Monte Carlo method to update the value function V( ). Specifically, according to the state Actual returns from start to finish To update V( ).

[0193] The policy π is updated through an optimization algorithm (such as gradient ascent) to select more valuable actions in future decisions. The goal is to minimize the negative expected return of the policy:

[0194]

[0195] Among them, θ is the strategy parameter.

[0196] The process of strategy execution, reward calculation, value function update, and strategy optimization is repeated until convergence or the predetermined number of iterations is reached.

[0197] After multiple rounds of iterations, the model learned how to generate ideas with higher innovation and practicality.

[0198] Input data set as initial state , in each step, the current state is generated through feature extraction and preprocessing .

[0199] By combining with information in long-term memory, the input data helps assess the novelty and usefulness of the innovative idea, thus influencing the reward function. Calculation.

[0200] Changes in input data will guide the adjustment of strategy π, which in turn affects the quality of generated ideas.

[0201] S330. Combine multiple types of data (such as text, images, videos, etc.) to support the generation of cross-modal innovation ideas. Specifically:

[0202] ① Obtain multimodal data from invention patents, including but not limited to text, images, and videos. For example, text materials such as technical documents, industry reports, and patent documents extracted from long-term memory; image materials related to specific technologies, such as product design drawings, experimental results charts, etc.; and related teaching videos, demonstration videos, etc.

[0203] ②Extract features from the data of each mode:

[0204] Text Data : Extract text features such as keywords, phrases, and topics.

[0205] Image data :Use computer vision technology to extract image features such as color, shape, and texture.

[0206] Video Data : Segment video frames, extract inter-frame difference features, audio features, etc.

[0207] ③ Mapping data of different modalities into a unified feature space for effective cross-modal fusion. This process can be achieved through a deep learning network, for example, using a multimodal self-attention mechanism to learn the correlation between features of different modalities.

[0208] ④ In a unified feature space, the feature vectors of different modes are , , The fusion is performed to form a comprehensive feature vector Z. The fusion method can be in the form of weighted summation, where the weight is determined by a specific function that takes into account the importance of each mode:

[0209]

[0210] in, , , are the weights of text, image, and video features respectively.

[0211] ⑤Use the fused feature vector Z as input to generate new cross-modal innovation ideas through the generative model. The generation process can follow the following formula:

[0212]

[0213] Where G is the generative model, is the feature representation of the generated innovative ideas, and θ is the model parameter.

[0214] ⑥ Evaluate the generated innovative ideas, including their novelty, practicality, and feasibility. The reinforcement learning framework mentioned above can be used to optimize the generation process, and the reward mechanism can be used to guide the model to generate higher quality innovative ideas.

[0215] In order to reflect the originality, a complex function F is introduced, which comprehensively considers the importance of each modal data and the interaction between them. Assume that there is a function F defined as follows:

[0216]

[0217] Where, Ω represents all possible modal feature spaces; , , Represent the weights of text, image, and video features respectively; ( ), ( ), ( ) represent the nonlinear transformation of text, image and video features respectively; g( , , ) represents an interaction function, which measures the mutual influence between different modal features.

[0218] The goal of this function is to capture the complex relationships between different modal features and summarize these relationships through integration operations to obtain a comprehensive evaluation score F.

[0219] S340. Use explainable AI techniques (such as causal reasoning, counterfactual explanations, etc.) to improve the credibility and verifiability of innovative ideas. Specifically:

[0220] ① Establish a group of agents composed of multiple virtual experts, each of which represents different professional knowledge and technical backgrounds. These agents will simulate the communication and collaboration process of experts in the real world.

[0221] ② Obtain multimodal data from invention patents and extract features from these data to form a feature vector F. is given an initial state vector , the state vector contains the relevant features extracted from the feature vector F.

[0222] ③ In order to simulate the interaction between intelligent agents, an undirected graph G=(V,E) is constructed, where the node V={ , ,..., } represents the set of intelligent agents, and the edge set E represents the interaction relationship between intelligent agents.

[0223] Defining the adjacency matrix ,in Representing an Agent and Agent If the connection strength between and If there is no direct connection between =0.

[0224] ④ Define the interaction rules and communication protocols between agents to simulate the information sharing and collaboration process between experts. Agents can share their state vectors by exchanging messages. And update its own state vector based on the received information.

[0225] Define the message passing function M( , ) is an intelligent agent Towards an Agent The method of sending messages, where and Agents and The state vector of .

[0226] ⑤Collaboration process:

[0227] a. Agents share their state vectors through a specific message passing mechanism For example, the agent You can use messages Towards an Agent Share their knowledge on key technical terms.

[0228]

[0229] b. The agent that receives the message will update its state vector according to the received information. For example, the agent Received from After receiving the message, it will update its state vector .

[0230]

[0231] Where α is the learning rate, which determines the influence of the received information.

[0232] ⑥ Promote the collision of diverse and creative thinking through continuous interaction between agents. Agents continuously update their state vectors, thereby generating new ideas and solutions. Define an innovation fusion function I( , ,..., ) to quantify the innovative ideas generated by the entire system, we can use principal component analysis (PCA) in multivariate statistical analysis to reduce dimensions and extract key information, and define p=PCA( , ,..., ) as a component of the innovation integration function.

[0233] ⑦ When the state vector of the agent When a certain stable state is reached, each agent can generate innovative ideas based on its own state vector. Using the generative models discussed previously (such as variational autoencoders VAE or generative adversarial networks GAN), each agent Generate an innovative idea .

[0234] ⑧Finally, all agents jointly evaluate the generated innovative ideas and provide feedback to further optimize and improve these ideas. Through collective evaluation, agents can identify the most promising ideas and feed them back to the entire system for subsequent optimization and implementation. A collective evaluation function C( , ,..., ) to assess the quality of innovative ideas and make adjustments based on the evaluation results.

[0235] Among them, F is the feature vector, which contains the key features extracted from the invention patent; For intelligent agents The state vector of the agent interaction network; G = (V, E) is an undirected graph, where V is a node set and E is an edge set; A is an adjacency matrix, which represents the connection strength between agents; M ( , ) is the message passing function, representing the agent Towards an Agent The message sent; α is the learning rate, which controls the influence of the message on the state update; p is the principal component vector, which is extracted from the state vector of the agent through PCA; For intelligent agents Generated innovative ideas; C( , ,..., ) is a collective evaluation function that evaluates the quality of innovative ideas generated by all agents.

[0236] S350, evaluate the novelty and practicality of the generated ideas to ensure that they meet the predetermined standards and goals. Specifically:

[0237] ① Based on the technical trends, key words and hot topics mentioned in the invention patent, define a set of evaluation criteria S={ , ,..., }, where each Represents a specific evaluation dimension. For example, novelty can be measured by the difference from existing knowledge, and practicality can be evaluated by the potential scope of application.

[0238] ②For each innovative idea , extract the relevant feature vector These eigenvectors Including but not limited to keyword frequency, technical relevance, potential application scenarios, etc.

[0239] For the convenience of comparison, the eigenvector Perform standardization to obtain the standardized feature vector .

[0240] ③ A comprehensive novelty function N(⋅) based on calculus equations and multiple regression equations is used to evaluate the novelty of each idea.

[0241]

[0242] Where,Ω represents the feature space; w( ) is the eigenvector Weight distribution in feature space Ω;

[0243] ( ) is the eigenvector Average weight distribution in the existing knowledge base.

[0244] ④ A practicality function U(⋅) based on a multiple regression equation is used to evaluate the practicality of each idea.

[0245]

[0246] in, ( ) is the eigenvector In the evaluation criteria The score under Is the evaluation criteria The weight coefficient of .

[0247] ⑤ A comprehensive evaluation function E(⋅) combining the novelty score and practicality score is used to comprehensively evaluate each idea.

[0248]

[0249] Where: γ is the weight coefficient of the novelty score.

[0250] ⑥ Set a threshold T, only when the comprehensive evaluation score E( ) is higher than the threshold T, the idea Only then will they be considered to have potential.

[0251]

[0252] ⑦ Ideas that will be selected Feedback is then fed back to the multi-agent system to further refine these ideas.

[0253] Based on the feedback results, the agent can update its state vector , thereby generating more optimized and practical ideas in the next round of iterations.

[0254] like Figure 5 As shown, step S400 at least includes steps S410-S460:

[0255] S410, collect user feedback on system output through questionnaire surveys, online evaluations, etc. Specifically:

[0256] ① Customized feedback questionnaires are designed based on the technical trends, key words and hot topics mentioned in the invention patents.

[0257] ② Send customized feedback questionnaires to users, who may be technical experts, researchers or ordinary users; the questionnaires can be distributed through various channels such as email, social media platforms, mobile applications, etc.

[0258] ③Integrate an online evaluation function on the front-end interface of the system, so that users can directly rate and comment on the ideas output by the system; the rating system adopts a five-star rating mechanism, allowing users to quickly give a comprehensive rating of the ideas; the comment area allows users to freely express their views and suggestions on a specific idea.

[0259] ④ The collected feedback data includes user ratings, comments, and completed questionnaire results; natural language processing technology is used to perform sentiment analysis on user comments to automatically identify positive or negative feedback; for open-ended questions in the questionnaire, natural language processing technology is also used for topic modeling to summarize the main opinions.

[0260] ⑤ Quantitatively analyze the collected feedback data and extract key indicators, such as novelty score, practicality score, relevance and coherence score, etc.; these indicators match the evaluation criteria defined in the innovation evaluation module mentioned above.

[0261] ⑥ Based on the results of quantitative analysis, a feedback report is compiled; the report should not only contain statistical data, but also include specific suggestions and opinions from users; through the feedback report, you can clearly understand which aspects have been recognized by users and which aspects still need improvement.

[0262] ⑦ Apply the feedback results to subsequent iterations of the system, especially to adjust the novelty and practicality evaluation criteria in the innovation evaluation module; the feedback results can also help improve the long-term memory and innovation generation modules, so that the system can better understand and generate truly novel and creative ideas.

[0263] S420. Analyze the output of the model and evaluate its accuracy and innovation. Specifically:

[0264] ① Obtain various forms of generated content from the output of the model, such as text descriptions, images, videos, etc.; these output contents reflect the ideas generated by the model based on the input data (key technical terms, hot topics in the field, emerging technology trends, etc.).

[0265] ② Use natural language processing technology to compare whether the terms in the model output are consistent with the key technical terms in the input data; evaluate whether the model correctly understands and uses these terms; use information retrieval and expert review to verify the accuracy and reliability of the text description generated by the model; for images and video materials, evaluate their accuracy through visual similarity analysis and expert review.

[0266] ③ The novelty function is used to evaluate the novelty of the model output ideas; the novelty function considers whether the idea goes beyond the existing knowledge and has unique contributions; the practicality function is used to evaluate the practical application value of the model output ideas; the practicality function considers whether the ideas can be applied in actual scenarios and the potential impact after application.

[0267] ④ For data in multiple modalities (text, image, video), adopt corresponding analysis methods respectively; text data: use semantic similarity analysis, topic model and other technologies; image data: apply computer vision technology, such as image classification, target detection, etc.; video data: combine audio and visual information, and use deep learning models for content analysis.

[0268] ⑤ Based on the above evaluation results, a comprehensive evaluation report is generated; the report includes the scores of various evaluation indicators, as well as a comprehensive evaluation of overall accuracy and innovation; the report may also put forward improvement suggestions for the model output for subsequent iterations.

[0269] S430, monitor the operating status of the system, including key indicators such as response time and resource utilization. Specifically:

[0270] ① Collect real-time data related to system operation, including but not limited to: response time: the time interval from when the model receives a request to when it returns a result; CPU utilization: the utilization rate of the server CPU; memory utilization: the ratio of currently used memory to total memory; disk I / O: the speed of disk read and write operations; network bandwidth: the speed of upstream and downstream network traffic; number of concurrent requests: the number of requests being processed at the same time point.

[0271] ② Based on the key technical terms in the input data, establish a fast search index to ensure that relevant terms and their context can be quickly located when processing requests; monitor the index building time and index size to ensure the efficiency and lightweight of the index.

[0272] ③ Monitor whether the system can respond quickly to new topics or hot events and generate relevant content; record the time interval from the emergence of hot topics to the generation of corresponding content by the model.

[0273] ④ Monitor the system’s ability to predict emerging technology trends, including prediction accuracy and prediction time window; track the time difference between the system’s prediction and actual development, as well as the deviation between the prediction results and the actual situation.

[0274] ⑤ Monitor the processing time of different types of data (text, images, videos); analyze the differences in the impact of different data types on system resources.

[0275] ⑥ Automatically adjust resource allocation based on monitored resource utilization, such as increasing or decreasing the number of server nodes; monitor the efficiency of the automatic adjustment mechanism to ensure timely response to load changes.

[0276] ⑦ Implement anomaly detection algorithms to identify possible problems in the system; such as sudden surges in CPU usage, memory leaks, and other abnormal situations, immediately triggering the alarm mechanism.

[0277] ⑧Generate performance monitoring reports regularly to summarize the system's operating status; the report includes trend charts of key performance indicators and a comprehensive analysis of the system's operating status; provide suggestions for improvement, such as optimizing resource allocation strategies and improving processing speed.

[0278] S440, automatically detect and repair errors or abnormal behaviors in the system. Specifically:

[0279] ① Carry out preliminary cleaning and verification of input data to ensure data quality; verify key technical terms; check whether the terms are complete, whether there are spelling errors or semantic ambiguities; confirm the matching degree of the terms with the corresponding fields.

[0280] ② Monitor key indicators during system operation, such as response time, resource utilization, etc.; Abnormal behavior detection: Use machine learning algorithms to analyze input data in real time and identify potential abnormal patterns; Detect whether there are abnormal values ​​in the input data within a specific time period, such as data mutations, unreasonable inputs, etc.

[0281] ③ Perform a logical consistency check on the output generated by the model to ensure that the output conforms to the expected behavior pattern; check whether the output content conflicts with the key technical terms in the input data; grammatical error identification: check the grammatical structure of the output to ensure the correctness of the language expression; detect grammatical errors in the output content, such as sentence structure errors, inconsistent tenses, etc.

[0282] ④ Record the specific information of all abnormal behaviors, including the time of occurrence, abnormal type, scope of impact, etc. Automatic correction attempt: For errors that can be automatically fixed, immediately apply the preset correction rules; for example, for simple grammatical errors, they can be corrected directly. Manual intervention request: For complex or errors that cannot be automatically fixed, generate a work order to notify maintenance personnel to intervene manually.

[0283] ⑤ Collect feedback information from user interactions, including user opinions, suggestions or error reports on the output content; pay special attention to user feedback on the novelty and creativity of the output content; Feedback analysis and application: analyze user feedback and identify common problems or patterns; apply feedback to the model training process to improve the quality and accuracy of future outputs.

[0284] ⑥ Regularly fine-tune the model based on anomaly detection results and user feedback; adjust the model's parameter settings to optimize its performance when processing specific types of data; Model version control: Implement a version control system to ensure that each update is clearly recorded; track the effects of each version so that you can roll back to the previous stable version when necessary.

[0285] ⑦ Automatically run tests after each code change to ensure that the new code does not introduce new errors; Automatic deployment: When the test passes, the updated model is automatically deployed to the production environment.

[0286] S450: Adjust the system based on feedback and analysis results, including model parameters, algorithm selection, etc. Specifically:

[0287] ① Collect user feedback on system output content, including but not limited to evaluation of novelty, creativity and practicality; user feedback may include direct evaluation, suggestions or error reports, etc. Feedback analysis: Use natural language processing technology to analyze user feedback and extract key information; analyze the consistency and novelty of technical terms, hot topics and trends mentioned in user feedback.

[0288] ② Collect various performance indicators during system operation, such as response time, throughput, accuracy, etc.; pay special attention to indicators related to long-term memory and innovation generation, such as the model's novelty score, creativity score, etc.; Performance analysis: perform statistical analysis on these performance indicators to find bottlenecks or anomalies; analyze the correlation between these indicators and user feedback to identify the main factors affecting user experience.

[0289] ③ Based on user feedback and performance indicators, evaluate the effectiveness of current model parameters; analyze which parameter settings help improve the novelty and creativity of the output; algorithm selection evaluation: examine the performance of different algorithms when processing different types of data (such as text, images); determine which algorithm combination can better capture the long-term memory characteristics in the input data.

[0290] ④ Adjust model parameters such as learning rate, weight decay, etc. according to the analysis results; optimize parameter configuration to improve the long-term memory capacity of the model and promote innovative generation algorithm selection: select an algorithm or algorithm combination that is more suitable for the current task requirements based on the evaluation results; consider using more advanced algorithms to enhance the creativity and adaptability of the model.

[0291] ⑤ Design an experimental plan to compare and test the adjusted system with the original system; verify the effectiveness of new parameters and algorithms to ensure that they have indeed improved system performance; effect evaluation: collect experimental data and conduct statistical tests to ensure that the improvement measures have significantly improved system performance; pay attention to the impact of the improvement measures on the novelty and creativity of the output content.

[0292] ⑥ Integrate proven parameter adjustments and algorithm selections into the system; ensure that the new version is compatible with the existing system and can transition smoothly; continuous monitoring: After deploying the new version, continue to monitor system performance and user feedback; be prepared to adjust the optimization strategy again based on new feedback and data at any time.

[0293] like Figure 6 FIG. 1 is a block diagram of a system for generating a scientific and technological innovation big data model according to an embodiment of the present application. The system may include:

[0294] The data collection and preprocessing module automatically acquires scientific research data from multiple sources, including but not limited to academic papers, patent documents, conference reports, open source projects, industry reports, etc.; it performs preprocessing operations such as cleaning, deduplication, and standardization on the collected data to improve data quality and reduce noise interference.

[0295] The feature extraction and knowledge graph construction module extracts key information and concepts from the text through natural language processing technologies such as word embedding, entity recognition, and relationship extraction; and uses the extracted information to construct a domain knowledge graph, including the relationship network between entities, attribute information, and their interactions, to form a structured knowledge representation.

[0296] The long-term memory and adaptive learning module stores and manages the structured data and knowledge graphs obtained from the data integration and knowledge graph construction module as part of the long-term memory; it continuously adjusts and optimizes the content of the long-term memory based on new data and knowledge, ensuring that the model can continue to improve with the development of time and new knowledge; it supports personalized learning paths for different users or application fields, and can adjust learning strategies and content according to specific needs.

[0297] The innovation generation and reasoning module performs reasoning and logical inference based on information in long-term memory to help generate new insights and hypotheses; identifies relevant knowledge and background information in specific situations to provide richer and more targeted support for innovation generation; regularly evaluates the effectiveness of long-term memory and learning strategies, and makes necessary adjustments based on the evaluation results.

[0298] The query and access service module provides interfaces and services to enable other modules to efficiently query and access information in the knowledge graph; implements a flexible and easy-to-use API interface, and supports multiple query methods, including but not limited to precise query, fuzzy query, and complex query (such as based on SPARQL syntax).

[0299] The performance monitoring and optimization module monitors the operating status of the system, including key indicators such as response time and resource utilization; automatically detects and repairs errors or abnormal behaviors in the system; and adjusts the system based on feedback and analysis results, including model parameters, algorithm selection and other aspects.

[0300] The following are its main beneficial effects:

[0301] (1) Comprehensive coverage of scientific research data: The present invention automatically obtains scientific research data from multiple sources, including academic papers, patent documents, conference reports, open source projects, industry reports, etc., ensuring the comprehensiveness and diversity of data, thereby providing rich and multi-faceted support for scientific research innovation.

[0302] (2) Efficient data preprocessing: By performing preprocessing operations such as cleaning, deduplication and standardization on the collected data, the data quality is improved, noise interference is reduced, and the accuracy of subsequent processing is ensured.

[0303] (3) Constructing high-quality knowledge graphs: Natural language processing techniques (such as word embedding, entity recognition, and relationship extraction) are used to extract key information from text and construct structured domain knowledge graphs, which help the model better understand the complex relationships and background knowledge in the scientific research field.

[0304] (4) Support for long-term memory and adaptive learning: By maintaining long-term memory and supporting adaptive learning mechanisms, the present invention can continuously adjust and optimize the content of long-term memory, ensuring that the model can continue to improve with the development of time and new knowledge, providing a solid foundation for generating innovative ideas.

[0305] (5) Generation of innovative ideas: We use long-term memory to generate innovative ideas and optimize the generation process through technologies such as reinforcement learning, thereby improving the quality and practicality of generated ideas and promoting the efficiency and effectiveness of scientific research innovation.

[0306] (6) Efficient query and access services: It provides a flexible and easy-to-use API interface that supports multiple query methods, enabling other modules to efficiently query and access information in the knowledge graph, thereby speeding up the acquisition and utilization of scientific research information.

[0307] (7) Continuous updating and optimization: The knowledge graph is updated regularly to reflect the latest research results and development trends, ensuring the timeliness and accuracy of the knowledge graph. At the same time, query performance and maintenance strategies are continuously optimized to ensure the stable operation of the system.

[0308] (8) Security and privacy protection: Security authentication and permission management functions are implemented to ensure that only authorized users or modules can access specific data, thus protecting sensitive information in the knowledge graph.

[0309] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for generating a scientific and technological innovation big data model based on a big model, characterized in that: include: S100, extract and organize raw data from various scientific research data sources, perform denoising and standardization, and build a graph structure containing domain knowledge; S200, integrates long-term memory module in the system, continuously updates and stores historical data, supports dynamically adjusted learning algorithms, and adapts to different data inputs; S300, based on the data in long-term memory, use the generative model to generate innovative ideas, and combine reinforcement learning technology to optimize the generation process and adjust the generation strategy through feedback; S400, collect user feedback information, conduct sentiment analysis and topic modeling, and optimize system parameters and algorithms; The S300 includes steps S310-S350: S310, using information in long-term memory to generate novel ideas and solutions through generative models; S320. Use reinforcement learning algorithms to guide the generation process of innovative ideas and encourage the model to explore more valuable and innovative solutions through reward mechanisms. S330. Combine multiple types of data to support cross-modal innovation idea generation, and introduce function F, which is defined as follows: ； Where, Ω represents all possible modal feature spaces; , , Represent the weights of text, image, and video features respectively; ( ), ( ), ( ) represent the nonlinear transformation of text, image and video features respectively; g( , , ) represents an interaction function that measures the mutual influence between different modal features; the goal of function F is to capture the complex relationship between different modal features and summarize these relationships through integral operations to obtain a comprehensive evaluation score F; S340. Use explainable artificial intelligence technology to improve the credibility and verifiability of innovative ideas; S350. Evaluate the novelty and practicality of generated ideas to ensure they meet predetermined standards and objectives.

2. The method for generating a big scientific and technological innovation data model based on a big model according to claim 1, characterized in that: Before the step of extracting and organizing the raw data, it also includes: Obtain key technical terms and hot topics in the field related to the scientific research data source.

3. The method for generating a big scientific and technological innovation data model based on a big model according to claim 1, characterized in that: The steps of constructing the graph structure of domain knowledge specifically include: Build an index of technical terms and monitor index build time and index size.

4. The method for generating a big scientific and technological innovation data model based on a big model according to claim 1, characterized in that: The integration steps of the long-term memory module specifically include: Historical data is categorized and stored, and a dynamically adjusted learning algorithm is set up.

5. The method for generating a big scientific and technological innovation data model based on a big model according to claim 1, characterized in that: The step of collecting user feedback information specifically includes: User feedback on the generated ideas was collected through questionnaires and online reviews.

6. The method for generating a big scientific and technological innovation data model based on a big model according to claim 5, characterized in that: The steps of sentiment analysis and topic modeling specifically include: Perform sentiment analysis on user reviews to identify positive and negative feedback and summarize opinion points.

7. The method for generating a big scientific and technological innovation data model based on a big model according to claim 1, characterized in that: The steps of optimizing system parameters and algorithms specifically include: The parameters of the long-term memory module and the generation model are adjusted according to user feedback and sentiment analysis results.

8. The method for generating a big scientific and technological innovation data model based on a big model according to claim 1, characterized in that: The optimization steps of the parameters and algorithms specifically include: Generate comprehensive evaluation reports regularly to analyze system operation status and performance indicators to guide subsequent system improvements.

Images