Scientific research achievement full-period management system and method based on multi-mode intelligent analysis

Abstract

The invention discloses a scientific research achievement full-period management system and method based on multi-mode intelligent analysis, and relates to the field of achievement collaborative management and knowledge enhancement of scientific research and test units. Through business process automation, knowledge enhancement retrieval and intelligent pushing, full-process intelligent management of scientific research achievements from generation, examination, submission, archiving to reutilization is realized, comprehensive application of achievements of scientific research and test units can be effectively promoted, and innovation vitality of scientific research personnel is better stimulated.

Description

Technical Field

[0001] The present invention relates to the field of collaborative management of scientific research results and knowledge enhancement in scientific research and experimental units, and specifically to a full-cycle management system and method for scientific research results based on multimodal intelligent analysis. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and may not constitute prior art.

[0003] Scientific research results (such as reports, standards, papers, patents, and systems) are the core assets of scientific research institutions and are of great value and significance to their academic and technological influence, resource acquisition and sustainable development, technology transformation and economic benefits, institutional reputation and talent attraction, strategic support, and social responsibility. However, scientific research institutions have long faced three major pain points in the management and utilization of scientific research results: First, the business management process is inefficient. The number, type, and status of scientific research results are unclear, information is fragmented, and the integration of heterogeneous data is difficult. This leads to repeated registration and reporting, duplication of effort by researchers (for example, the patent application process and the expense reimbursement process require the submission of the same materials repeatedly), and repeated verification by agency managers, resulting in low efficiency and prone to errors.

[0004] Second, there is a lack of knowledge transformation. Research results data is managed in a decentralized manner, archived in an irregular manner, and lacks a unified knowledge management platform and intelligent retrieval and analysis tools. This prevents the effective transformation and accumulation of single, fragmented research results into systematic, reusable knowledge assets that can be inherited and reused by the institution, making it difficult to support the shared application of domain expertise.

[0005] Third, there's insufficient integration and application of knowledge. Research institutions accumulate a vast and growing body of research knowledge, including reports, standards, papers, and regulations, stored as static files. This knowledge is essential for researchers to carry out their work in an orderly manner. On the one hand, it requires researchers to expend time and effort to search and organize. On the other hand, it's not integrated with existing research-related systems and processes, leaving it disconnected from the work environment. This leads to low knowledge utilization and a waste of resources.

[0006] The above-mentioned problems involve the main links in the generation, archiving and reuse of scientific research results, which influence and restrict each other: inefficient processes cause scientific researchers to spend a lot of energy on repeated filling and manual coordination, rather than core scientific research activities; knowledge fragmentation hinders the accumulation and inheritance of professional knowledge in the field, making it difficult to transform historical results into reusable intelligent assets; application disconnection makes the knowledge base a static archive, unable to be deeply integrated with business scenarios such as experimental design and project application, forming a "knowledge island". These problems and their intertwined impacts not only restrict scientific research efficiency and results conversion rate, but also weaken the competitiveness of scientific research units in complex scientific research tasks (such as experiments and multidisciplinary collaborative innovation). There is an urgent need for a full-link, dynamically adaptive intelligent management solution for scientific research results to achieve a fundamental transformation of scientific research results from "distributed storage" to "intelligent services".

[0007] According to the existing relevant literature search, there are few public documents at home and abroad on the whole process of intelligent management of scientific research results within scientific research institutions from generation, review, submission, archiving to reuse.

[0008] Judging from these public reports, the existing research literature mainly has limitations such as single data management integration (most systems only focus on the management of single-type results such as patents and papers, and lack unified integration of multimodal data such as test reports and project documents), weak cross-system collaboration (approval processes are isolated from business systems such as financial reimbursement and confidentiality review, and data synchronization relies on manual operations), low knowledge reuse rate (search results are poorly matched with user needs, and the citation rate of historical results in existing systems is low), and insufficient scenario adaptability (mostly adopting a passive service model, relying on users to actively initiate queries, and unable to predict scientific research needs, such as actively pushing risk warnings at the project startup stage). These limitations cannot meet the business needs of scientific research and experimental units in terms of knowledge management, job training, knowledge application, and operational guidance. There are no public reports on a full life cycle management system for scientific research results that is suitable for scientific research and experimental units. Summary of the Invention

[0009] The purpose of the present invention is to provide a full-cycle management system and method for scientific research results based on multimodal intelligent analysis to address the problems of inefficient process control, unclear results base, lack of knowledge transformation, and broken knowledge circulation in the management of scientific research results in current scientific research and experimental units.

[0010] Specifically, the present invention aims to solve the following technical problems: First, there are issues of duplication of effort and manual errors caused by inefficient business processes; The second is the low conversion efficiency caused by the decentralized management of scientific research results; The third is the problem of insufficient utilization caused by the disconnection between static knowledge storage and business scenarios.

[0011] Specifically, in response to the above-mentioned problems, the present invention proposes a full-cycle management system and method for scientific research results based on multimodal intelligent analysis. Through business process automation, knowledge-enhanced retrieval and intelligent push, it realizes the full-process intelligent management of scientific research results from generation, review, submission, archiving to reuse, which can effectively promote the comprehensive application of the results of scientific research and experimental units and better stimulate the innovative vitality of scientific researchers.

[0012] The technical solutions of the present invention are as follows: A full-cycle management system for scientific research results based on multimodal intelligent analysis, including: The business control module is used to automate the conditional-triggered multi-department collaborative approval process through weighted state transition equations and a dynamic role weight matrix; The knowledge enhancement module, based on retrieval enhancement generation technology, performs semantic analysis and intelligent transformation of multimodal scientific research results to generate structured knowledge assets; The precise service module uses collaborative filtering algorithms and dynamic weight adaptation mechanisms to achieve intelligent push of scientific research results and service closed-loop optimization.

[0013] Furthermore, the business management and control module includes: a process modeling unit configured to define process nodes, roles, and rules using a business process model and symbols, and to describe process state transitions using a weighted state transition equation and a dynamic role weight matrix; a role modeling unit configured to generate a dynamic role weight matrix based on the number of inter-department collaborations and approval efficiency; The automated trigger unit is configured to calculate the trigger probability by multiplying the role weight by the state transition condition. When the calculation result is greater than or equal to a preset threshold, the automatic flow is triggered, otherwise it is transferred to manual processing.

[0014] Furthermore, the weighted state transfer equation is as follows:

[0015] in, is the current state, For the process from the current state Enter the next state No. Conditional rules, For the The role weight of each condition, To trigger an event, is the urgency coefficient; The dynamic role weight matrix is ​​as follows:

[0016] in, For the department With the department The collaboration weight between The degree of interdepartmental collaboration is rare. For the department With the department The relative frequency of collaboration, The number of approval processes completed per unit time.

[0017] Furthermore, the trigger probability is expressed as:

[0018] in, For the current state and roles Under the condition, transfer to the next state If the result is 1, it means that the transfer is bound to occur and the process is allowed to be triggered automatically. If the result is 0, it means that the transfer is impossible. Intermediate values ​​mean that the transfer may occur, but there are probability restrictions or role weight adjustments, which require manual operation. From the state Spontaneous transition to state The basic probability of Indicates that the transfer path is allowed. Indicates that the transfer path is prohibited; Representing a role Transfer Path The influence weight of .

[0019] Furthermore, the knowledge enhancement module includes: A multimodal processing unit configured to perform structured parsing of scientific research results in Excel, PDF, and Word formats. Excel parsing maps cell data to structured objects, PDF parsing extracts logical structures through optical character recognition and layout analysis, and Word parsing extracts hierarchical elements of paragraphs and tables. The hybrid retrieval unit is configured to use a semantic encoding model to generate document vectors and perform linear weighted retrieval in combination with keyword relevance scores.

[0020] Furthermore, the hybrid retrieval unit performs the following operations: The document content is semantically encoded through a dual encoder model to generate a dense vector representing the semantic information; Calculate keyword weight scores by counting word frequency and inverse document frequency; The semantic search results and keyword search results are fused through the weight coefficient. The calculation formula is: semantic weight coefficient × semantic similarity score + (1-semantic weight coefficient) × keyword relevance score.

[0021] Furthermore, the knowledge enhancement module also includes: The fusion unit is configured to calculate the semantic similarity between the retrieved fragment and the user query, assign fusion weights based on the temperature parameter, and realize information reorganization through the attention mechanism; The generation unit is configured to generate structured text based on the pre-trained language model and perform grammatical correction and logical consistency optimization.

[0022] Furthermore, the precise service module includes: a dynamic synchronization unit configured to filter a candidate document set based on a semantic vector similarity threshold; A collaborative filtering unit configured to calculate user similarity and construct a neighbor set based on user behavior data; The scenario adaptation unit is configured to perform weighted filtering on the data pool according to the business link type.

[0023] Furthermore, the precise service module also includes: Push calculation unit, configured to integrate semantic similarity, user similarity, scenario weight and initial weight coefficient of user information selected for business link Generate TOP-N push sequence; Real-time feedback unit, configured to dynamically adjust weight coefficients based on user click-through rates .

[0024] The present invention also proposes a full-cycle management method for scientific research results based on multimodal intelligent analysis, which includes: The dynamic jump and automatic triggering of multi-department collaborative approval processes are achieved through the collaborative calculation of weighted state transfer equations and dynamic role weight matrices; Perform optical character recognition, semantic coding, and hybrid retrieval on multimodal scientific research results to generate a structured knowledge base; Optimize collaborative filtering algorithms based on user behavior feedback data to achieve dynamic adaptation of knowledge push and business scenarios.

[0025] Compared with the existing technology, the beneficial effects of the present invention are: 1. Improved process efficiency: The system eliminates repetitive work for researchers in areas such as achievement reporting and fund reimbursement through intelligent process automation and knowledge reuse.

[0026] 2. Systematization of knowledge assets: Transform scattered scientific research results into a searchable and relatable knowledge base, for example, mapping wind tunnel test parameters, paper conclusions, and patent technology points into a unified entity relationship; through intelligent analysis of user behavior (such as frequently queried literature and frequently cited test databases), identify common needs in the field and automatically archive them as a "knowledge template library", making implicit knowledge explicit.

[0027] 3. Improved user experience and satisfaction: Accurate knowledge delivery is achieved. For example, when a user initiates the "Acoustic Wind Tunnel Test Design" task, the system automatically delivers abnormal data reports from related historical tests, patents for similar scenarios, and other relevant information. New employees can use the system to quickly retrieve best practice documents and historical problem solutions in the field, reducing their learning curve. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 Data flow diagram for the knowledge enhancement module; Figure 2 Filter flowchart for scene data pool. DETAILED DESCRIPTION

[0029] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus comprising the element.

[0030] The features and performance of the present invention are further described in detail below with reference to the embodiments.

[0031] Example 1 A full-cycle management system for scientific research results based on multimodal intelligent analysis, including: The business control module is used to automate the condition-triggered multi-department collaborative approval process through weighted state transition equations and a dynamic role weight matrix. It should be noted that the function of the business control module is described as: enabling multi-department collaborative approval based on a workflow engine and supporting condition-triggered process automation. The knowledge enhancement module, based on retrieval enhancement generation technology, performs semantic analysis and intelligent transformation of multimodal scientific research results to generate structured knowledge assets. It should be noted that the function of the knowledge enhancement module is described as follows: with the semantic analysis and intelligent transformation of scientific research results as the core goal, through RAG retrieval enhancement generation (retrieval-generation-fusion) technology, it transforms scattered, unstructured, multimodal scientific research results (such as reports, papers, patents, standards, etc.) into reusable and relatable structured knowledge assets, and supports the dynamic generation of text content that meets user needs, thereby realizing the structured storage and intelligent generation of scientific research results. The precise service module realizes the intelligent push of scientific research results and service closed-loop optimization through collaborative filtering algorithms and dynamic weight adaptation mechanisms. It should be noted that the functional description of the precise service module is: establishing a seamless connection with the knowledge enhancement module, using the document vector library formed by the above-mentioned BGE model as the semantic basis of the push system, and dynamically adjusting the weight according to the user scenario, performing push calculations, triggering two-way optimization according to user behavior (such as click-through rate, etc.), and forming a collaborative evolution mechanism of knowledge update → scenario adaptation → push calculation → parameter optimization → closed-loop feedback.

[0032] In this embodiment, specifically, the business control module includes: A process modeling unit is configured to define process nodes, roles, and rules using BPMN (Business Process Model and Notation), and describe process state transitions using weighted state transition equations and a dynamic role weight matrix; a role modeling unit configured to generate a dynamic role weight matrix based on the number of inter-department collaborations and approval efficiency; The automated trigger unit is configured to calculate the trigger probability by multiplying the role weight by the state transition condition. When the calculation result is greater than or equal to a preset threshold, the automatic flow is triggered, otherwise it is transferred to manual processing.

[0033] In this embodiment, specifically, the weighted state transfer equation is as follows:

[0034] in: is the current state; For the process from the current state Enter the next state No. Conditional rules (e.g. “Only when the results pass the review can we enter the approval stage”); For the The role weight of each condition (such as file integrity weight , approval level weight ); A trigger event (such as "automatically flow to the next node" or "trigger financial review"); is the urgency factor (such as urgent tasks , common tasks ); The dynamic role weight matrix based on user behavior is as follows:

[0035] in: For the department With the department The collaboration weight between them; The rarity of collaborative relationships between departments; For the department With the department the relative frequency of collaboration; The number of approval processes completed per unit time.

[0036] In this example, specifically, the rarity of interdepartmental collaboration Calculated by the following formula:

[0037] in: is the total number of collaborations between all department pairs; It is a department With the department The number of times collaboration occurs between The constant 1 is used to avoid the denominator being zero.

[0038] In this embodiment, specifically, the department With the department Relative frequency of collaboration Calculated by the following formula:

[0039] in: is the total number of departments.

[0040] In this embodiment, specifically, the number of processes completed for approval per unit time is Calculated by the following formula:

[0041] in: It is a department With the department exist The number of collaborative approvals completed within a certain period of time.

[0042] In this embodiment, specifically, the trigger probability is expressed as:

[0043] in: For the current state and roles Under the condition, transfer to the next state If the result is 1, it means that the transfer is bound to occur and the process is allowed to be triggered automatically. If the result is 0, it means that the transfer is impossible. Intermediate values ​​mean that the transfer may occur, but there are probability restrictions or role weight adjustments, which require manual operation. From the state Spontaneous transition to state The basic probability of Indicates that the transfer path is allowed. Indicates that the transfer path is prohibited; Representing a role Transfer Path The influence weight of .

[0044] In this embodiment, it should also be noted that after each task is completed, according to the process efficiency (such as completion time t) to dynamically adjust the weight:

[0045] in: For the The weight after the condition is updated; is the learning efficiency (e.g. 0.1); is the process efficiency of the current task, which may be a metric such as completion time or error rate; Represents the historical expected value of process efficiency, i.e., the average value; Indicates the variance of process efficiency, used for normalization changes.

[0046] In this embodiment, if Figure 1 As shown, the knowledge enhancement module includes: A multimodal processing unit configured to perform structured parsing of scientific research results in Excel, PDF, and Word formats. Excel parsing maps cell data to structured objects, PDF parsing extracts logical structures through optical character recognition and layout analysis, and Word parsing extracts hierarchical elements of paragraphs and tables. The hybrid retrieval unit is configured to use a semantic encoding model to generate document vectors and perform linear weighted retrieval in combination with keyword relevance scores.

[0047] In this embodiment, it should be noted that the processing goal of the multimodal processing unit is to perform a multimodal unified representation of the input results (text, charts, formulas, etc.), as follows: a. Excel data analysis Use openpyxl to parse Excel data, extract the questions, options, and answers in the cells, map each row of data into a structured object containing complete question information, and finally format the structured data into text paragraphs; b. PDF data analysis When processing PDF data, OCR (Optical Character Recognition) is first required to convert text within the image into editable text. Layout analysis identifies the text layout within the PDF file, including the structure of lines, paragraphs, and pages. This helps understand the logical structure of the text, enabling more accurate information extraction.

[0048] c. Word data analysis Use python-docx to parse paragraphs, tables, images, and other elements in Word documents. Clean and convert text formats (e.g., remove redundant symbols, unify encodings, etc.). Save the parsed data to a target format (e.g., JSON, text files, or database records).

[0049] In this embodiment, specifically, the hybrid retrieval unit performs the following operations: The document content is semantically encoded through a dual encoder model to generate a dense vector representing the semantic information; Calculate keyword weight scores by counting word frequency and inverse document frequency; The semantic search results and keyword search results are fused through the weight coefficient. The calculation formula is: semantic weight coefficient × semantic similarity score + (1-semantic weight coefficient) × keyword relevance score.

[0050] In this embodiment, it should be noted that the goal of the hybrid search unit is to efficiently retrieve semantic information related to user needs from massive heterogeneous scientific research results based on a hybrid search of BGE semantic search and BM25 keyword search, as follows: a. BGE model text representation The BGE model is used to semantically encode the document content and generate a dense vector representation:

[0051] in: To input original documents of scientific research results (such as scientific research reports, patent abstracts, papers, etc.); It is a dual encoder model used to map text into dense semantic vectors, such as the BERT-based architecture; The vector dimension (such as 768, 1024, etc.) output by the model is used to control the fineness of the semantic representation. Each dimension of this vector corresponds to a semantic direction captured by the model, which is used to describe the semantic information, topic, entity or syntactic structure of the document.

[0052] b. BM25 keyword extraction Use the BM25 algorithm to extract keywords, and calculate keyword weights by counting document word frequency TF and inverse document frequency IDF:

[0053] in: represents the relevance score of word t (a word in the query) to query q in document d; Measure the importance of word t in the corpus; Indicates the number of times word t appears in the document; parameter Control the saturation of word frequency TF; parameter Adjust document length normalization strength; is the document length; is the average document length in the corpus.

[0054] c. Collaboration between semantic retrieval and keyword retrieval Through linear weighting, the advantages and disadvantages of semantic search (capturing deep semantic associations but potentially lacking keyword sensitivity) and keyword search (rapidly matching explicit keywords but failing to understand implicit semantics) are balanced, achieving synergy between BGE semantic search and BM25 keyword search.

[0055] in: is the semantic weight (usually set to 0.7), balancing the contribution of semantic retrieval and keyword retrieval; Based on the comprehensive score Sort the documents and prioritize returning the top-k document fragments with high semantic similarity and matching keywords.

[0056] In this embodiment, specifically, the knowledge enhancement module further includes: The fusion unit is configured to calculate the semantic similarity between the retrieved fragments and the user query, assign fusion weights based on the temperature parameter, and implement information reorganization through the attention mechanism. It should be noted that the fusion unit's fusion goal is to semantically align the retrieved fragmented information with the user's original input context and generate contextualized and structured new input for the generative model to use; The generation unit is configured to generate structured text based on the pre-trained language model and perform grammatical correction and logical consistency optimization. It should be noted that the goal of the generation unit is to generate text content (such as experimental reports, abstracts, etc.) that conforms to scientific research standards and is semantically coherent based on the fused input.

[0057] In this embodiment, it should be noted that the fusion unit is implemented by the following specific algorithm: a. Context similarity calculation For each search fragment Calculate the matching degree with the user query q and calculate the retrieval fragment through the vector generated by the BGE model And the semantic similarity of query q:

[0058] in: Represents a search fragment The semantic similarity with query q; Represents a search fragment dense semantic vector of ; represents the dense semantic vector of query q; Indicates the corrected search fragment length; is the length penalty coefficient (e.g. = -0.2) to avoid long texts dominating the results.

[0059] b. Fusion weight allocation Assign fusion weights based on similarity and information relevance:

[0060] in: is the fusion weight, based on similarity Dynamic allocation; is the temperature parameter, which controls the concentration of weight distribution. When it is larger (e.g. 1.0), it emphasizes generalized fusion and all fragments have similar weights; When it is smaller (for example, 0.5), it emphasizes exact matching, and the weight of high-similarity segments is significantly higher than that of low-similarity segments. Represents a search fragment The semantic similarity with query q; c. Information Reorganization Rules according to Perform weighted integration of fragmented information while retaining the key semantics of the original input:

[0061] in: Represents semantic fusion based on attention mechanism (such as Transformer's Cross-Attention); Can be a fragment The text content, vector representation, or structured information containing metadata such as location and timestamp.

[0062] In this embodiment, it should be noted that the generation unit specifically has the following functions: Text planning: Developing the structure and content outline of the generated text based on the fused input; Language generation: Use pre-trained text generation models, such as GPT, to generate text according to the planned structure; during the generation process, the model will consider grammar, semantics, and contextual coherence; during the generation process, through the multi-head self-attention mechanism Capture long-range dependencies and correct ambiguities:

[0063] in: It represents the target word generated by the model at step t, which is output after linear transformation of the weighted value vector; Indicates the current target word The semantic query vector is derived from the linear transformation of the input text; A key vector representing the input text, used to calculate relevance with the query vector; Represents the query vector With key vector The vector dimension of is used for normalization calculation; The value vector representing the input text carries the semantic information of the original text and is weighted by the attention weight to generate the output.

[0064] Post-processing: The generated text may need to be post-processed, including correcting grammatical errors, eliminating duplicate content, ensuring transaction consistency, etc.

[0065] Output: The final generated text is output to the user.

[0066] In this embodiment, specifically, the precise service module includes: a dynamic synchronization unit configured to filter a candidate document set based on a semantic vector similarity threshold; A collaborative filtering unit configured to calculate user similarity and construct a neighbor set based on user behavior data; A scenario adaptation unit is configured to perform weighted filtering on the data pool according to the type of business link; Push calculation unit, configured to integrate semantic similarity, user similarity, scenario weight and initial weight coefficient of user information selected for business link Generate TOP-N push sequence; Real-time feedback unit, configured to dynamically adjust weight coefficients based on user click-through rates .

[0067] In this embodiment, it should be noted that the dynamic synchronization unit is mainly used to achieve dynamic knowledge weight synchronization, as follows: The BGE model based on the knowledge enhancement module generates semantic vectors for scientific research documents and user queries, and uses similarity calculation results to screen the preliminary candidate document set, ensuring that the push calculation unit always uses the latest semantic vector representation generated by the knowledge enhancement module;

[0068] in: Represents the cosine similarity of the semantic vector generated by BGE model encoding; Represents the semantic vector generated by encoding the input document d through the BGE model; Represents the semantic vector generated by encoding the user query q through the BGE model.

[0069] The screening threshold is The candidate document set , and sort by similarity:

[0070] in: Represents the input document.

[0071] In this embodiment, it should be noted that the collaborative filtering unit is mainly used to implement push based on collaborative filtering, as follows: We use user ratings to assess user similarities and then make recommendations based on these similarities. For example, if user u has previously searched for knowledge findings from a research project and is highly similar to user v, we can predict that user v is also interested in searching for the same research findings. Therefore, we can push relevant research findings to user v.

[0072] user , The content similarity is calculated as follows:

[0073] in: Represents a user , similarity of content; Represents the set of knowledge items that user u participates in querying; Represents the set of knowledge items that user v participates in querying.

[0074] The above algorithm can be used to calculate the similarity between the current user's search content and other users, and the n users with the highest similarity to the current user are listed as the neighbor set. The collaborative filtering algorithm recommends the existing knowledge model of the users in the neighbor set to the current user, that is, the user's potential knowledge model; Based on the calculated similarity, the weighted average method is used to predict the probability of the current user u's demand for unsearched knowledge. The prediction table formula is:

[0075] in: represents the set of probabilities of the current user u’s demand for unsearched knowledge predicted by the weighted average method; Represents the current user Similar user sets; Represents the search frequency of similar user v for knowledge i.

[0076] In this embodiment, it should be noted that the scene adaptation unit is mainly used for weight adaptation of scene data pool filtering, as follows: After calculating the initial push knowledge, before pushing, in order to improve the push efficiency and accuracy, the data pool needs to be screened according to different business links. Each user corresponding to each business scenario has a set of initial data pool weights. The system group S represents the user's preference for each data pool data, so as to achieve more effective personalized knowledge push to users. Figure 2 shown.

[0077] Taking the project management system as an example, the project management system provides users with a help interface and pushes knowledge as the main input and output. The probability array P obtained after the knowledge array is filtered through the data pool can be expressed as:

[0078] in: Respectively represent the weight coefficients of user u in data pool 1, data pool 2 and data pool n; Represent the corresponding knowledge in data pool 1, data pool 2 and data pool n respectively.

[0079] In this embodiment, it should be noted that the push calculation unit is mainly used to implement TOP-N push and dynamic weight correction, as follows: The algorithm weight coefficient is used to calculate the candidate knowledge set. Finally, the initial weight coefficient is selected based on the corresponding user information of the business link to fuse the candidate knowledge set to calculate the probability of each knowledge push, and the push order is obtained from this. The corresponding ranking can be obtained based on the order of each pushed knowledge, and then the final push value is calculated based on the weight to obtain the recommended ranking. The specific formula is as follows:

[0080] Where, Indicates the weight of the ranking value based on the collaborative filtering algorithm; Indicates the weight of the ranking value based on the similarity of the candidate document set; Indicates the recommended ranking value based on the collaborative filtering algorithm; Represents the recommended ranking value based on the similarity of the candidate document set; Indicates the recommended ranking value of the mixed ranking calculated by weight. Finally, sort in descending order according to the probability set, and take The knowledge corresponding to the top N probabilities is pushed.

[0081] In this embodiment, it should be noted that the main function of the real-time feedback unit is to feed back user behavior to the knowledge enhancement module and update the semantic vector, as follows: Dynamic weight correction is to continuously optimize the push effect and enhance the system's adaptability by collecting user feedback.

[0082]

[0083] in: is the content similarity weight coefficient before iteration; is the updated weight coefficient after correction by user feedback; Is the learning coefficient, which controls the step size of weight correction; Is the cross entropy loss function Loss weight coefficient gradient;

[0084] The cross-entropy loss function is used to quantify the error between the push results and the user's actual behavior (such as clicks or no clicks).

[0085] in: A binary label indicating whether the user clicked on document d (1 = clicked, 0 = not clicked); The system predicts the user click probability and uses the sigmoid function to score Mapped to the range 0,1.

[0086] This embodiment also proposes a full-cycle management method for scientific research results based on multimodal intelligent analysis. The full-cycle management system for scientific research results based on multimodal intelligent analysis includes: The dynamic jump and automatic triggering of multi-department collaborative approval processes are achieved through the collaborative calculation of weighted state transfer equations and dynamic role weight matrices; Perform optical character recognition, semantic coding, and hybrid retrieval on multimodal scientific research results to generate a structured knowledge base; Optimize collaborative filtering algorithms based on user behavior feedback data to achieve dynamic adaptation of knowledge push and business scenarios.

[0087] Example 2 Example 2 takes a patent application as an example to further illustrate a full-cycle management system and method for scientific research results based on multimodal intelligent analysis proposed in Example 1, covering the entire process from generation, archiving to reuse, as follows: 1. Patent application stage Business control module: After the user submits the patent draft, the system automatically triggers the approval process: Call the rule engine to dynamically generate approval nodes (such as technical review, financial review, etc.) based on patent type (such as invention, utility model, etc.) and budget status.

[0088] Push tasks to corresponding departments (such as the technical department and the finance department) through the BPMN engine, and automatically extract keywords from patent abstracts (BM25 algorithm) to generate classification labels.

[0089] After approval, the system automatically generates a reimbursement form and associates the patent number to avoid repeated submission of materials.

[0090] Knowledge enhancement module: OCR scans patent documents, extracts metadata such as titles and abstracts, generates semantic vectors using the BGE model, and stores them in the knowledge base.

[0091] Intelligently fill in repetitive content (such as claim templates) to reduce manual writing time.

[0092] 2. Results archiving stage Knowledge enhancement module: The archived patent documents are associated with historical data, and a summary of technical points is automatically generated through the RAG generation phase to form structured knowledge entries.

[0093] Establish a mapping relationship between patents and wind tunnel test data (such as patent number → test number) to support cross-system retrieval.

[0094] 3. Knowledge reuse stage Precision service module: When a user logs in to the system, the system calls the hybrid push algorithm: Content matching: Based on the user's current query (such as "acoustic wind tunnel design"), the semantic similarity with the patents in the knowledge base is calculated .

[0095] Collaborative filtering: Analyze user historical behavior (such as browsing patents related to "aeroacoustics") and recommend knowledge items in similar fields.

[0096] Weighted ranking: comprehensive score , push TOP-5 related patents and technical reports.

[0097] Based on feedback such as users clicking on a "certain patent details page" or reading for more than 2 minutes, the (alpha) value is dynamically adjusted (such as increasing the collaborative filtering weight).

[0098] 4. Full life cycle closed loop The system continuously collects user behavior data and optimizes the push algorithm.

[0099] Periodically reconstruct historical achievements into knowledge graphs to enhance the relevance of domain expertise.

[0100] The above-described embodiments merely represent specific implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of protection of the present application. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the technical concept of the present application, and all such variations and improvements fall within the scope of protection of the present application.

[0101] This background section is provided to generally present the context of the invention, and the work of the presently named inventors, the work to the extent described in this background section, and aspects of the description in this section that did not constitute prior art at the time of filing are neither explicitly nor implicitly admitted to be prior art to the present invention.

Claims

1. A full-cycle management system for scientific research results based on multimodal intelligent analysis, characterized in that: include: The business control module is used to automate the conditional-triggered multi-department collaborative approval process through weighted state transition equations and a dynamic role weight matrix; The knowledge enhancement module, based on retrieval enhancement generation technology, performs semantic analysis and intelligent transformation of multimodal scientific research results to generate structured knowledge assets; The precise service module, through collaborative filtering algorithm and dynamic weight adaptation mechanism, forms a collaborative evolution mechanism of knowledge update → scenario adaptation → push calculation → parameter optimization → closed-loop feedback.

2. The scientific research achievement full-cycle management system based on multimodal intelligent analysis according to claim 1 is characterized in that: The business control module includes: a process modeling unit configured to define process nodes, roles, and rules using a business process model and symbols, and to describe process state transitions using a weighted state transition equation and a dynamic role weight matrix; a role modeling unit configured to generate a dynamic role weight matrix based on the number of inter-department collaborations and approval efficiency; The automated trigger unit is configured to calculate the trigger probability by multiplying the role weight by the state transition condition. When the calculation result is greater than or equal to a preset threshold, the automatic flow is triggered, otherwise it is transferred to manual processing.

3. The scientific research achievement full-cycle management system based on multimodal intelligent analysis according to claim 2 is characterized in that: The weighted state transfer equation is as follows: in, is the current state, For the process from the current state Enter the next state No. Conditional rules, For the The role weight of each condition, To trigger an event, is the urgency coefficient; The dynamic role weight matrix is ​​as follows: in, For the department With the department The collaboration weight between The degree of interdepartmental collaboration is rare. For the department With the department The relative frequency of collaboration, The number of approval processes completed per unit time.

4. The full-cycle management system for scientific research achievements based on multimodal intelligent analysis according to claim 3 is characterized in that: The trigger probability is expressed as: in, For the current state and roles Under the condition, transfer to the next state If the result is 1, it means that the transfer is bound to occur and the process is allowed to be triggered automatically. If the result is 0, it means that the transfer is impossible. An intermediate value means that the transfer may occur, but there are probability restrictions or role weight adjustments, which require manual operation. From the state Spontaneous transition to state The basic probability of Indicates that the transfer path is allowed. Indicates that the transfer path is prohibited; Representing a role Transfer Path The influence weight of .

5. The full-cycle management system for scientific research achievements based on multimodal intelligent analysis according to claim 2 is characterized in that: Knowledge enhancement modules include: A multimodal processing unit configured to perform structured parsing of scientific research results in Excel, PDF, and Word formats. Excel parsing maps cell data to structured objects, PDF parsing extracts logical structures through optical character recognition and layout analysis, and Word parsing extracts hierarchical elements of paragraphs and tables. The hybrid retrieval unit is configured to use a semantic encoding model to generate document vectors and perform linear weighted retrieval in combination with keyword relevance scores.

6. The full-cycle management system for scientific research achievements based on multimodal intelligent analysis according to claim 5 is characterized in that: The hybrid retrieval unit performs the following operations: The document content is semantically encoded through a dual encoder model to generate a dense vector representing the semantic information; Calculate keyword weight scores by counting word frequency and inverse document frequency; The semantic search results and keyword search results are fused through the weight coefficient. The calculation formula is: semantic weight coefficient × semantic similarity score + (1-semantic weight coefficient) × keyword relevance score.

7. The full-cycle management system for scientific research achievements based on multimodal intelligent analysis according to claim 6 is characterized in that: The knowledge enhancement module also includes: The fusion unit is configured to calculate the semantic similarity between the retrieved fragment and the user query, assign fusion weights based on the temperature parameter, and realize information reorganization through the attention mechanism; The generation unit is configured to generate structured text based on the pre-trained language model and perform grammatical correction and logical consistency optimization.

8. The full-cycle management system for scientific research achievements based on multimodal intelligent analysis according to claim 7 is characterized in that: The precise service module includes: a dynamic synchronization unit configured to filter a candidate document set based on a semantic vector similarity threshold; A collaborative filtering unit configured to calculate user similarity and construct a neighbor set based on user behavior data; The scenario adaptation unit is configured to perform weighted filtering on the data pool according to the business link type.

9. The full-cycle management system for scientific research achievements based on multimodal intelligent analysis according to claim 8 is characterized in that: The precise service module also includes: Push calculation unit, configured to integrate semantic similarity, user similarity, scenario weight and initial weight coefficient of user information selected for business link Generate TOP-N push sequence; Real-time feedback unit, configured to dynamically adjust weight coefficients based on user click-through rates .

10. A full-cycle management method for scientific research results based on multimodal intelligent analysis, characterized in that: A full-cycle management system for scientific research achievements based on multimodal intelligent analysis according to any one of claims 1 to 9, comprising: The dynamic jump and automatic triggering of multi-department collaborative approval processes are achieved through the collaborative calculation of weighted state transfer equations and dynamic role weight matrices; Perform optical character recognition, semantic coding, and hybrid retrieval on multimodal scientific research results to generate a structured knowledge base; Optimize collaborative filtering algorithms based on user behavior feedback data to achieve dynamic adaptation of knowledge push and business scenarios.

Images