A patent creative intelligent screening and evaluation method and system based on multi-agent cooperation

Abstract

The present application relates to the technical field of intelligent evaluation, and more particularly to a patent creative intelligent screening and evaluation method and system based on multi-agent cooperation, comprising: S1: receiving a patent creative text by a creative intelligent agent and performing semantic analysis, extracting a technical application scenario, an existing scheme pain point, a creative description and a core innovation point; S2: calling a semantic retrieval engine by a retrieval intelligent agent, performing parallel retrieval in multiple databases, and generating a retrieval result set; S3: performing evaluation reasoning by an evaluation intelligent agent from multiple dimensions of novelty, practicality, technical effect, scene difference degree and technical path coverage, and generating an evaluation conclusion; S4: feeding back the evaluation conclusion to a management system by a cooperative intelligent agent and pushing an evaluation notification; S5: monitoring new creations and triggering cyclic execution by the cooperative intelligent agent according to a preset scheduling rule. The present application realizes full-process automatic closed-loop processing, evaluation standardization and conclusion explainability, and significantly improves the patent creative screening efficiency and management cooperation.

Description

Technical Field

[0001] This invention relates to the field of intelligent evaluation technology, and in particular to a method and device for intelligent screening and evaluation of patented ideas based on multi-agent collaboration. Background Technology

[0002] With the rapid development of artificial intelligence technology, patent idea management is showing a significant trend towards intelligence and automation. Against this backdrop, to improve the efficiency of patent idea screening and the quality of evaluation, traditional techniques typically employ manual review, single search tools, and general process management platforms.

[0003] Manual review relies on personal experience to judge novelty and assess value. Because it's entirely manual, it suffers from long processing times and high labor costs when dealing with a large volume of ideas. Furthermore, inconsistent judgment standards among reviewers lead to highly subjective conclusions, increasing the risk of missed or incorrect screenings. This, to some extent, restricts the responsiveness of companies' patent strategy. Single-search tools perform patent novelty searches based on keyword matching, returning a list of relevant patents after the user inputs a search query. However, keyword searches cannot accurately capture the semantic meaning of technical solutions, resulting in numerous noisy documents and a high rate of missed detections. These tools only output search reports, lacking customized assessment capabilities for initial screening of ideas within companies. R&D personnel often struggle to understand the search results, requiring secondary interpretation by intellectual property specialists, leading to high communication costs. While general process management platforms digitize idea submission and approval processes, their core functions are limited to process recording and status tracking, lacking intelligent analysis, automatic retrieval, and quantitative assessment capabilities. After idea submission, manual processing is still required for each item, failing to create a complete closed-loop process from idea analysis, multi-database retrieval, comprehensive evaluation to conclusion feedback and automatic notification.

[0004] Furthermore, the aforementioned technical methods all rely on manual judgment for evaluation, employing fixed standards to assess patent ideas without considering the differentiated requirements for novelty and utility across different technical fields and application scenarios. In practical applications, a company's patent ideas involve multiple technical directions, and the requirements for evaluation accuracy differ significantly between core technologies and peripheral improvement technologies. While core technologies are fewer in number, they require in-depth searching and multi-dimensional evaluation; whereas most peripheral improvement technologies have relatively lower requirements for evaluation depth. However, once the traditional evaluation model is established, it is difficult to adjust. All ideas are processed at a uniform depth, leading to over-allocation of evaluation resources and further exacerbating the waste of human and time costs, resulting in significant shortcomings in terms of economy and practicality. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by providing a method for intelligent screening and evaluation of patent ideas based on multi-agent collaboration, comprising the following steps: S1: The creative intelligent agent receives the patent creative text submitted by the user, performs semantic analysis on the patent creative text, and extracts the technical application scenarios, pain points of existing solutions, creative descriptions and core innovations; S2: The retrieval agent calls the semantic retrieval engine based on the technology application scenario and the core innovation point to perform parallel retrieval in the patent database, academic literature database and technology report database, and generate a retrieval result set; S3: The evaluation agent evaluates and reasones the patent idea from multiple dimensions, including novelty, practicality, technical effect, scenario difference, and technical path coverage, based on the search result set, and generates an evaluation conclusion. S4: The collaborative intelligent agent fills the evaluation conclusion back into the corresponding record in the preset management system and sends the evaluation notification through the preset notification channel; S5: The collaborative intelligent agent monitors the submission of new ideas according to the preset scheduling rules, and triggers steps S1 to S4 to be executed in a loop when an idea to be evaluated is detected. S6: The creative agent, retrieval agent, evaluation agent, and collaborative agent interact with each other through a preset multi-agent communication protocol to arrange messages, schedule tasks, and synchronize states, so as to achieve collaborative execution of patent idea screening and evaluation processing.

[0006] Preferably, in step S1, semantic parsing of the patent idea text includes: The patent creative text is used to identify intent based on a pre-trained language model in order to determine the core appeal of the creative idea. Technical features are extracted from the patent creative text based on named entity recognition technology to obtain technical entity phrases. Based on semantic similarity, the innovative points of the technical entity phrases are clustered, and the unstructured creative text is transformed into structured technical application scenarios, pain points of existing solutions, creative descriptions and core innovative points; The semantic similarity is calculated in the following way: ,in, and These represent two technical entity phrases, and These represent the corresponding semantic vector representations; when When the similarity exceeds a preset threshold, the corresponding technical entity phrases will be grouped into the same innovation point cluster.

[0007] Preferably, in step S2, the semantic retrieval engine is invoked to perform multi-database parallel retrieval, including: A hybrid retrieval strategy is constructed based on the application scenarios of the technology and the core innovations, employing retrieval enhancement generation technology. The hybrid retrieval strategy combines vector retrieval and keyword retrieval, and performs semantic matching simultaneously in patent databases, academic literature databases, and technical report databases. The matching results returned by each database are aggregated and sorted based on relevance scores to generate a search result set containing highly relevant patents and documents. Among them, the relevance score Calculate using the following formula: ,in, The semantic similarity score for vector retrieval. This is the matching score for keyword retrieval. The weighting coefficient has a range of values. This is used to adjust the weight ratio of semantic matching and exact matching in the final ranking.

[0008] Preferably, in step S3, the multi-dimensional evaluation reasoning includes: The core innovation points are compared with the existing technical features in the search results set to evaluate the coverage of technical paths; The scenario difference between the patented idea and existing technologies is calculated, and a quantitative score is given by combining practicality indicators and technical effect parameters; wherein, the scenario difference... Calculate using the following formula: ,in, This represents the scene semantic vector corresponding to the patented idea. This represents the scene semantic vector corresponding to the existing technology; Based on the combined novelty evaluation results and quantitative scoring results, a comprehensive evaluation score is calculated, matched against preset evaluation rules, and an assessment conclusion including an evaluation level and explanatory reasons is generated; the comprehensive evaluation score Calculate using the following formula: ,in, The score represents the novelty factor. Indicates the practicality score. Indicates the score for technical effectiveness. Indicates the coverage of technology paths. to These are the weight coefficients for the corresponding dimensions, and they satisfy... .

[0009] Preferably, the evaluation conclusion is generated, including: The evaluation conclusion is generated based on the joint judgment result of whether the core innovation points are covered by existing technologies and the degree of difference in scenarios, combined with the comprehensive evaluation score. The joint determination result is obtained by calculating a joint determination value. Confirmed, the calculation formula is: ,in, This is an adjustment coefficient, and its value range is... ; when If the score exceeds the preset coverage threshold, it is determined that the core innovation points are partially covered or highly related. If the comprehensive evaluation score is lower than the first preset threshold, an evaluation conclusion that the application is not recommended is generated. If the comprehensive evaluation score is between the first and second preset thresholds, an evaluation conclusion that the application should be adjusted and resubmitted is generated, and the differentiated details or technical adjustment directions that need to be clarified are specified in the explanatory reasons. when If the difference is less than the preset difference threshold and the comprehensive evaluation score is higher than the second preset threshold, it is determined that the core innovation point is not covered and the difference is significant. An evaluation conclusion recommending the application is generated, and the basis supporting the novelty of the core innovation point is output.

[0010] Preferably, in step S4, the evaluation conclusion is backed up to the preset management system and a notification is pushed, including: The collaborative agent analyzes the evaluation conclusions and extracts the evaluation level, explanatory reasons, and associated original creative identifiers; Based on the original creative identifier, the evaluation level and explanatory reasons are automatically written into the designated field of the corresponding record in the management system to form a data closed loop of creative submission, evaluation processing and result retention; Simultaneously, a message push event is triggered, the target notification channel is determined according to system configuration or user preferences, the evaluation level and explanatory reasons are encapsulated into a message format adapted to the target notification channel, and the corresponding interface is called to complete the targeted message push; Among them, the push priority of the k-th notification channel Determine using the following formula: ,in, Indicates the historical success rate of reaching an agreement. Indicates timeliness score, Indicates user preference weights, , , Let be the weight coefficient, and satisfy... .

[0011] Preferably, in step S5, automatically monitoring new creative submissions and triggering cyclic execution according to preset scheduling rules includes: The collaborative intelligent agent initiates a monitoring task according to a preset scheduling cycle, scanning the creative submission records in the management system whose status is marked as new, pending processing, or unevaluated. After a new creative idea is detected, the corresponding creative text is automatically extracted, and the creative agent, retrieval agent, evaluation agent and collaborative agent are triggered in sequence according to the trigger priority of the record to be processed to execute the processing flow. Among them, the triggering priority Calculate using the following formula: ,in, This indicates the duration of the record waiting time. Indicates the urgency of the business to which the creative idea belongs. Indicates the submitter's priority. , , These are the weighting coefficients.

[0012] Preferably, it also includes responding to the intelligent agent's assisted review comments: The responding agent sends a supplementary search request to the searching agent through the multi-agent communication protocol; The search agent re-executes a multi-database search based on the technical disputes and returns the supplemented relevant comparative literature to the response agent. Based on the supplementary comparative literature and the original patent idea, the response agent generates a rebuttal strategy, key points of the response, and directions for modifying the application documents in response to examination inquiries, and pushes them to the intellectual property specialist. The technical dispute matching degree G is calculated according to the following formula: ,in, This indicates the number of feature terms in the supplementary comparative literature that match the technical dispute. This represents the total number of features of the technical dispute.

[0013] Preferably, it also includes safety control steps: When data is transmitted between the intelligent agents, the transmitted patent idea text and evaluation data are encrypted. Users who access and operate the management system shall be subject to hierarchical access verification to restrict unauthorized access. Before data backfilling and notification push, sensitive information in the text is anonymized; the anonymization process uses a character masking function, defined as: ,in, For the text field to be processed, Indicates reserved fields The former The character and the following Each character, with the middle part replaced by a preset mask character; Record the entire operation log and perform data leakage prevention audits, and calculate the anomaly score of the operation behavior. : ,in, This is the current operation record. For the first Anomaly detection feature functions, These are the weight coefficients of the corresponding feature functions. The total number of anomaly detection features; when An anti-leakage alarm is triggered when the value exceeds the preset alarm threshold.

[0014] Based on the same concept, this invention also provides a patent idea intelligent screening and evaluation system based on multi-agent collaboration, including: The creative semantic parsing module is used by the creative intelligent agent to receive the patent creative text submitted by the user, perform semantic parsing on the patent creative text, and extract the technical application scenarios, pain points of existing solutions, creative descriptions and core innovations. The multi-database parallel retrieval module is used by the retrieval agent to call the semantic retrieval engine based on the technology application scenario and the core innovation point, and to perform parallel retrieval in the patent database, academic literature database and technology report database to generate a retrieval result set; The multi-dimensional evaluation and reasoning module is used by the evaluation agent to evaluate and reason about the patent idea from multiple dimensions, including novelty, practicality, technical effect, scenario difference, and technical path coverage, based on the search result set, and generate an evaluation conclusion. The evaluation conclusion collaborative execution module is used by the collaborative intelligent agent to backfill the evaluation conclusion into the corresponding record in the preset management system and send the evaluation notification through the preset notification channel. The timed monitoring and cyclic scheduling module is used by the collaborative intelligent agent to monitor the submission of new ideas according to preset scheduling rules, and to trigger the cyclic execution of the idea semantic parsing module, the multi-database parallel retrieval module, the multi-dimensional evaluation reasoning module and the evaluation conclusion collaborative execution module when an idea to be evaluated is detected. The multi-agent communication and collaboration module is used for message interaction, task orchestration, and state synchronization between the creative agent, retrieval agent, evaluation agent, and collaboration agent through a preset multi-agent communication protocol, so as to realize the collaborative execution of patent idea screening and evaluation processing.

[0015] Compared with the prior art, the beneficial effects of the present invention are: This invention achieves fully automated closed-loop processing of patent ideas, from submission, parsing, retrieval, and evaluation, through a creative intelligent agent that automatically receives and semantically parses patent idea text, a retrieval intelligent agent that searches in parallel across multiple databases, an evaluation intelligent agent that performs quantitative evaluation and reasoning from multiple dimensions, and a collaborative intelligent agent that automatically fills in the conclusions and pushes notifications. Compared to the traditional manual review model, the processing time for a single idea is reduced from days to minutes, manpower input is reduced, and the efficiency of patent idea screening and scalable processing capabilities are significantly improved. This invention employs quantitative evaluation and reasoning of intelligent agents across five dimensions: novelty, practicality, technical effectiveness, scenario differentiation, and technical path coverage. It generates assessment conclusions that include evaluation levels and explanatory justifications, achieving standardized evaluation criteria and traceable conclusions. Researchers can directly understand the assessment conclusions and reasoning without the need for secondary interpretation by intellectual property specialists, effectively reducing communication costs and avoiding the problems of inconsistent standards and highly subjective conclusions inherent in traditional manual evaluations. This invention utilizes a collaborative intelligent agent to automatically populate the evaluation conclusions into the corresponding records in a pre-defined management system and pushes evaluation notifications through pre-defined notification channels, thus achieving a data closed loop of creative submission, evaluation processing, and result retention. Simultaneously, the collaborative intelligent agent automatically monitors new creative submissions according to pre-defined scheduling rules and triggers cyclical execution, achieving unattended automated processing, preventing creative omissions, and improving collaborative efficiency.

[0016] This invention achieves loosely coupled collaboration among creative agents, retrieval agents, evaluation agents, and collaborative agents through a pre-defined multi-agent communication protocol for message interaction, task orchestration, and state synchronization. When the function of a certain agent needs to be upgraded or replaced, only adjustments need to be made according to the communication protocol interface, without affecting the normal operation of other agents. The system has high flexibility and scalability, and can adapt to the patent idea management needs of enterprises of different sizes. Attached Figure Description

[0017] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0018] Figure 1 This is a flowchart of a patent idea intelligent screening and evaluation method based on multi-agent collaboration according to the present invention; Figure 2 This is a flowchart of the semantic parsing process for the creative intelligent agent of this invention; Figure 3 This is a flowchart of the multi-database parallel retrieval process of the retrieval agent in this invention; Figure 4 This is a flowchart illustrating the multi-dimensional evaluation reasoning process of the intelligent agent in this invention. Figure 5 This is a flowchart illustrating the execution process of the collaborative intelligent agent in this invention. Figure 6 This is a timing diagram for the timed monitoring loop execution of the present invention; Figure 7 This is a flowchart illustrating the intelligent agent assistance mechanism of the present invention. Figure 8 This is a schematic diagram of the overall system architecture of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. Obviously, the described embodiments are only some, not all, of the embodiments described in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without creative effort are within the scope of protection of this application.

[0020] Those skilled in the art will understand that, unless otherwise stated, the singular forms “a” and “an” used herein, and “the”, may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0021] First Embodiment Please see Figure 1 As shown in the figure, this embodiment provides a method for intelligent screening and evaluation of patent ideas based on multi-agent collaboration, which includes the following steps: S1: The creative agent receives the patent creative text submitted by the user, performs semantic analysis on the patent creative text, and extracts the technical application scenarios, pain points of existing solutions, creative descriptions and core innovations.

[0022] Specifically, users input unstructured creative text through the front-end interface: In warehouse AGV navigation, existing solutions use LiDAR, which is costly and prone to losing positioning among complex shelves (pain point). This creative idea proposes a low-cost navigation solution based on the fusion of visual landmarks and inertial measurement units (IMU) (creative description). The core lies in the design of a dynamic visual landmark encoding algorithm and a tightly coupled visual / IMU data model (core innovation), applied to warehousing and logistics scenarios (application scenario).

[0023] Please see Figure 2 As shown, in step S1, semantic parsing of the patent idea text includes: This embodiment uses a pre-trained language model to identify the intent of patent creative text in order to determine the core appeal of the creative idea. Specifically, the RoBERTa-base model is used as the pre-trained language model, which contains a 12-layer Transformer encoder and a hidden layer dimension of 768. The model is pre-trained in a domain-adaptive manner on a patent domain corpus and fine-tuned on an annotated creative intent classification dataset. It can identify intent categories such as proposing new solutions, improving existing technologies, and combining applications. Through model inference, the core appeal of this creative text is identified as proposing a new navigation solution to reduce costs and improve robustness. Technical features are extracted from patent creative text using Named Entity Recognition (NER) technology to obtain technical entity phrases. Specifically, in this embodiment, a BiLSTM-CRF sequence labeling model is used for named entity recognition. The input is a token sequence after WordPiece word segmentation, and the output is an entity label for each token (using BIO labeling mode). The defined entity categories include: technical field entities (TECH-FIELD), hardware component entities (HW-COMP), algorithm model entities (ALG-MODEL), and performance indicator entities (PERF-METRIC). After model inference, the extraction results are: technical field entities (AGV vehicles, warehousing and logistics), hardware component entities (LiDAR, visual landmarks, IMU), algorithm model entities (dynamic visual landmark encoding algorithm, data tightly coupled model), and performance indicator entities (cost, positioning loss, robustness). Based on semantic similarity, we cluster innovative points of technical entity phrases and transform unstructured creative texts into structured technical application scenarios, pain points of existing solutions, creative descriptions and core innovative points. Semantic similarity is calculated as follows: ,in, and These represent two technical entity phrases (e.g., visual landmarks and dynamic visual landmark encoding algorithms). and These represent the corresponding semantic vector representations; when When the similarity exceeds a preset threshold, the corresponding technical entity phrases are grouped into the same innovation point cluster. Specifically, in this embodiment, the semantic vector is obtained by averaging the hidden states of the last layer of the pre-trained BERT model, resulting in a vector dimension of 768. For multi-word entities, the hidden vectors of each token in the entity are first averaged, and then L2 normalization is performed. The calculated result is... =Greater than the preset similarity threshold of 0.7, so it is classified into the same innovation point cluster "visual navigation scheme". Similarly, the similarity between IMU and data tightly coupled model is 0.82, so it is classified into the cluster "multi-sensor fusion". Finally, the unstructured text is transformed into structured text: technical application scenario (warehouse logistics AGV navigation), pain points of existing solutions (high cost of lidar, easy loss of positioning between complex shelves), creative description (low-cost navigation scheme based on visual landmark and IMU fusion), core innovation points ((1) dynamic visual landmark encoding algorithm; (2) visual / IMU data tightly coupled model).

[0024] S2: The retrieval agent calls the semantic retrieval engine based on the technology application scenario and core innovation points to perform parallel retrieval in the patent database, academic literature database and technology report database, and generate a retrieval result set.

[0025] Please see Figure 3 As shown, in step S2, the semantic retrieval engine is invoked to perform multi-database parallel retrieval, including: This paper employs Retrieval Enhancement Generation (RAG) technology to construct a hybrid retrieval strategy based on the application scenarios and core innovations of the technologies. Specifically, in this embodiment, "warehousing and logistics AGV navigation" is used as the scenario query, and "dynamic visual landmark encoding algorithm" and "visual / IMU data tightly coupled model" are used as technology queries to construct a hybrid retrieval strategy. Large Language Modeling (LLM) is used to automatically generate expanded search expressions based on the scenario query and the technology query. For the technology query "visual / IMU data tightly coupled model," LLM generates expanded synonyms including "visual inertial odometry," "VIO tightly coupled," and "multi-sensor fusion positioning," and constructs a Boolean search expression: (visual OR visual inertial) AND (IMU OR inertial measurement) AND (tightly coupled OR fusion) AND (navigation OR positioning). The expanded search expression serves as the input for keyword retrieval, while the semantic vector of the original technology query serves as the input for vector retrieval. The system simultaneously performs vector retrieval (based on the Milvus vector database engine, using the IVF_FLAT index, with cluster center number nlist=1024) and keyword retrieval (based on the Elasticsearch full-text search engine, using the BM25 algorithm to calculate text matching scores) in three databases. A hybrid retrieval strategy combining vector retrieval and keyword retrieval is implemented, and semantic matching is performed simultaneously in patent databases, academic literature databases, and technical report databases.

[0026] The matching results returned by each database are aggregated and sorted based on relevance scores to generate a search result set containing highly relevant patents and documents.

[0027] Among them, relevance score Calculate using the following formula: ,in, The semantic similarity score for vector retrieval (normalized to the [0,1] interval by Min-Max). The BM25 match score for keyword retrieval (also processed by Min-Max normalization). The weighting coefficient has a range of values. This is used to adjust the weight ratio of semantic matching and exact matching in the final ranking. Specifically, in this embodiment, considering the professional nature of terminology and the diversity of synonyms in patent documents, semantic matching is more likely to discover substantially relevant prior art. Therefore, it is set to... This allows semantic similarity to play a dominant role in the final ranking. For example, in a patent document about "visual-inertial SLAM," the vector retrieval score... Keyword search score Then the final score System Filtering The records are used to generate a search result set that includes highly relevant patents (such as a visual navigation patent CN2023XXXXXX) and academic literature (such as a research paper on visual inertial odometry).

[0028] S3: The evaluation agent evaluates and reasones on the patent idea from multiple dimensions, including novelty, practicality, technical effect, scenario difference and technical path coverage, based on the search result set, and generates an evaluation conclusion.

[0029] Please see Figure 4 As shown, in step S3, the multi-dimensional evaluation reasoning includes: To assess the coverage of the technology path, the core innovation points are compared with the existing technical features in the search results set. Specifically, in this embodiment, the evaluation agent first uses a feature comparison module to compare the core innovation points "dynamic visual landmark encoding algorithm" and "visual / IMU data tight coupling model" with the technical features disclosed in each document in the search results set. The comparison revealed that existing technologies include visual navigation schemes based on fixed QR code landmarks (covering the "visual landmark" feature), but do not mention dynamic encoding strategies; and loosely coupled visual and IMU schemes (covering the "visual + IMU" combination), but do not disclose the tight coupling fusion model. Based on the comparison results, the Jaccard similarity coefficient is used to calculate the coverage of the technology path. The total number of technical features of the core innovation points is 5 (dynamic encoding, landmark generation strategy, visual feature extraction, IMU pre-integration, and tight coupling optimization), of which 1.5 features are disclosed in existing technologies (1 for fully disclosed visual feature extraction and 0.5 for partially disclosed visual + IMU combination). The technology path coverage is then calculated. ; The scenario difference between the patent idea and existing technology is calculated, and a quantitative score is generated by combining practicality indicators and technical effect parameters; among them, the scenario difference... Calculate using the following formula: ,in, This represents the semantic vector of the scenario corresponding to the patent idea. This represents the scene semantic vector corresponding to the existing technology. Specifically, in this embodiment, This is the semantic vector for the complex warehouse shelves in this creative scenario. To expand the semantic vector of indoor warehousing in existing technology scenarios, the following calculations were performed. (The difference is significant). Combined with other dimensions for scoring: Novelty (The core innovations have not been fully disclosed), practicality (The market demand for AGVs in warehousing and logistics is large, and the solution is feasible based on commercial-grade sensors.) Technical effects ; Based on the combined results of the novelty evaluation and the quantitative scoring, a comprehensive evaluation score is calculated, matched against preset evaluation rules, and an assessment conclusion including the evaluation level and explanatory reasons is generated; the comprehensive evaluation score... Calculate using the following formula: ,in, The score represents the novelty factor. Indicates the practicality score. Indicates the score for technical effectiveness. Indicates the coverage of technology paths. to These are the weight coefficients for the corresponding dimensions, and they satisfy... Specifically, in this embodiment, considering that the most crucial dimensions of patent examination are novelty and inventiveness (among which scenario difference and technical path coverage together constitute key indicators of inventiveness), weights are set... , , , , Substituting the data yields .

[0030] Preferably, the evaluation conclusion is generated, including: The evaluation conclusion is generated based on the joint judgment result of whether the core innovation points are covered by existing technologies and the degree of difference in scenarios, combined with the comprehensive evaluation score. The joint determination result is obtained by calculating the joint determination value. Confirmed, the calculation formula is: ,in, This is an adjustment coefficient, and its value range is... , This is used to adjust the relative importance of technology path coverage and scenario differences in joint determination. When At that time, the judgment focuses more on the coverage of technical features; when In this case, the determination focuses more on the degree of difference at the application scenario level. This embodiment sets... This indicates that technical feature coverage and scenario differences have equal weight in the determination. Substituting... , Calculated ; when If the score exceeds the preset coverage threshold, it is determined that the core innovation points are partially covered or highly related. If the comprehensive evaluation score is lower than the first preset threshold, an evaluation conclusion that the application is not recommended is generated. If the comprehensive evaluation score is between the first and second preset thresholds, an evaluation conclusion that the application should be adjusted and resubmitted is generated, and the differentiated details or technical adjustment directions that need to be clarified are specified in the explanatory reasons. when When the difference is less than a preset difference threshold and the overall evaluation score is higher than a second preset threshold, it is determined that the core innovation point is not covered and the difference is significant. An evaluation conclusion recommending the application is generated, and the basis supporting the novelty of the core innovation point is output. Specifically, in this embodiment, the preset coverage threshold... Preset difference threshold First preset threshold The second preset threshold These thresholds were determined based on statistical analysis of historical granted patent cases: when and At that time, the authorization rate in historical cases reached over 92%. Because... (Difference threshold), and (Second preset threshold) The system determines that the core innovation points are not covered and the scene differences are significant, and generates the evaluation conclusion of "recommended application", and outputs the following in the explanatory reasons: "The existing technology does not cover the dynamic visual landmark coding and visual / IMU tight coupling features, and the application scenarios are significantly different in complex occluded storage environments. It is recommended to apply for a tight coupling model as an independent claim and the dynamic coding algorithm as a subordinate technical feature."

[0031] S4: The collaborative intelligent agent fills the evaluation conclusion back into the corresponding record in the preset management system and sends the evaluation notification through the preset notification channel.

[0032] Please see Figure 5 As shown, in step S4, the evaluation conclusion is populated back into the preset management system and a notification is pushed, including: The collaborative intelligent agent analyzes the evaluation conclusion, extracts the evaluation level, explanatory reasons, and associated original creative identifier. Specifically, in this embodiment, the recommended application level and explanatory reasons are extracted and automatically written into the corresponding data table row of the management system based on the original creative identifier (ID: AGV-2023-01), forming a closed loop. Simultaneously, a message push event is triggered. Based on the original creative identifier, the evaluation level and explanatory reasons are automatically written into the designated fields of the corresponding record in the management system to form a data closed loop of creative submission, evaluation processing and result retention; Simultaneously, a message push event is triggered, the target notification channel is determined based on system configuration or user preferences, the evaluation level and explanatory reasons are encapsulated into a message format adapted to the target notification channel, and the corresponding interface is called to complete the targeted message push; Among them, the push priority of the k-th notification channel Determine using the following formula: ,in, Indicates the historical success rate of reaching an agreement. Indicates timeliness score, Indicates user preference weights, , , Let be the weight coefficient, and satisfy... .

[0033] Specifically, before pushing the notification, the system determines the target notification channel and calculates the push priority for Lark and email. Regarding Lark channels: historical reach success rate Timeliness (Real-time), User Preferences ;set up , , Calculated Similarly, calculate the email... .because The system selects Lark as the target push channel, encapsulates the message into a Lark card message format (including the title "Creative Evaluation Completed", evaluation level label, explanatory text, and original creative link), and calls the Lark Open Platform Robot API to complete the targeted push to the submitting user and intellectual property specialist. S5: The collaborative intelligent agent monitors the submission of new ideas according to the preset scheduling rules, and triggers steps S1 to S4 to be executed in a loop when an idea to be evaluated is detected.

[0034] Please see Figure 6 As shown, in step S5, automatically monitoring new creative submissions and triggering cyclic execution according to preset scheduling rules includes: The collaborative intelligent agent starts monitoring tasks according to a preset scheduling cycle, scanning and managing creative submission records in the system whose status is marked as new, pending, or unevaluated. After a new creative idea is detected, the corresponding creative text is automatically extracted, and the creative agent, retrieval agent, evaluation agent and collaborative agent are triggered in sequence according to the trigger priority of the record to be processed to execute the processing flow. Among them, trigger priority Calculate using the following formula: ,in, This indicates the duration of the record waiting time. Indicates the urgency of the business to which the creative idea belongs. Indicates the submitter's priority. , , These are the weighting coefficients.

[0035] For example, record 1 waiting time (Normalized to 0.8), Business urgency (Core Project), Submitter Priority (Ordinary Engineer); Record 2 , , (Submitted by the Director). The system sets [specific parameters] based on the current strategy's emphasis on business urgency. , , Calculated , The system prioritizes extracting the text from record 1 to trigger process S1, achieving unattended closed-loop scheduling.

[0036] S6: The creative agent, retrieval agent, evaluation agent, and collaborative agent communicate, arrange tasks, and synchronize states through a preset multi-agent communication protocol to achieve collaborative execution of patent idea screening and evaluation processing.

[0037] In this embodiment, the multi-agent communication protocol is implemented based on the gRPC framework and the Protocol Buffers serialization protocol. The protocol defines the following message types: Task Request message: contains Task ID, Initiating agent ID, Target agent ID, Task type enumeration value, Payload data (JSON format), and Timestamp.

[0038] Task response message (TaskResponse): contains task ID, status code (success / failure / processing), payload data, and error information.

[0039] StatusSync message: contains agent ID, current status (idle / busy / error), current task ID, and heartbeat timestamp.

[0040] Each agent registers its own service with the service registry (implemented based on etcd) upon startup and obtains the network addresses of other agents through the service discovery mechanism. Heartbeat messages are sent every 5 seconds; if no heartbeat is received for three consecutive times, the agent is considered offline, and the collaborating agents suspend task distribution to that agent until it recovers. This loosely coupled design allows each agent to be upgraded and scaled independently. For example, during peak idea submission periods, the number of instances of the retrieval agent can be increased to improve concurrent processing capabilities.

[0041] Please see Figure 7 As shown, it also includes responses to comments from intelligent agents assisting in the review process: The responding agent sends a supplementary retrieval request to the retrieving agent via a multi-agent communication protocol. The retrieval agent re-executes a multi-database search based on the technical disputes and returns the supplementary relevant comparative literature to the response agent; Based on supplementary comparative literature and the original patent concept, the response agent generates argumentation strategies, key points of responses, and directions for revising application documents in response to examination inquiries, and pushes them to the intellectual property specialist. The technical dispute matching degree G is calculated according to the following formula: ,in, This indicates the number of feature terms in the supplementary comparative literature that match the technical disputes. This represents the total number of features in the technical dispute. Specifically, in this embodiment, the examination opinion states that the patent application lacks inventiveness compared to a prior art document (D1), and the technical dispute is the nonlinear mapping feature of the dynamic visual landmark encoding algorithm. The responding agent sends a supplementary search request to the searching agent via a multi-agent communication protocol. The searching agent re-searches based on this dispute and returns three supplementary prior art documents, representing the total number of features in the dispute. (Including: nonlinear mapping, landmark encoding length, fault tolerance, number of iterations, decoding strategy), supplementing the number of matched feature terms in the literature. (Only mentioning fault tolerance), then The extremely low matching degree indicates that neither D1 nor existing literature provides technical insights into this dispute. Therefore, the response agent generates the following argument strategy: emphasizing that D1 does not disclose nonlinear mapping features and that those skilled in the art have no incentive to combine them, and generating key points of the response and directions for modifying the application documents (adding nonlinear mapping features to the exclusive claim), which are then submitted to the intellectual property specialist for review.

[0042] Preferably, it also includes safety control steps: When data is transmitted between intelligent agents, the transmitted patent idea text and evaluation data are encrypted. Specifically, in this embodiment, the data transmission between the creative agent, the retrieval agent, the evaluation agent, and the collaborating agent uses the TLS 1.3 protocol for transport layer encryption, and the ECDHE key exchange algorithm is used during the TLS handshake phase to achieve forward security. Simultaneously, the transmitted patent idea text payload is encrypted using application layer AES-256-GCM, and the session key is rotated periodically through a key management service (KMS) with a rotation cycle of 24 hours. Even if the transport layer is intercepted, attackers cannot decrypt the creative content. Users accessing and operating the management system are subject to hierarchical permission verification to restrict unauthorized access. Specifically, in this embodiment, users accessing the management system must undergo hierarchical permission verification using RBAC (Role-Based Access Control). The roles and corresponding permissions are defined as follows: R&D personnel role: can only submit ideas and view the ideas and evaluation results submitted by themselves; Department head role: can view the ideas and evaluation results of all members in their department; Intellectual property specialist role: can view and manage all ideas and perform review and rejection operations of evaluation results; System administrator role: can configure system parameters and user permissions, but cannot view the original text of the ideas (separate permissions). Each API call is authenticated and permission verified through JSON Web Token (JWT). Before data backfilling and notification push, sensitive information in the text is anonymized; the anonymization process uses a character masking function, defined as: ,in, For the text field to be processed, Indicates reserved fields The former The character and the following The characters are replaced with a preset mask character in the middle. Specifically, in this embodiment, the identification of sensitive fields is based on a preset sensitive word rule library and a regular expression matching engine. The types of sensitive fields covered include: internal project codes, undisclosed customer names, inventor ID numbers, contact information, etc. For example, when pushing out notifications, the sensitive field "Customer Name: Nebula Logistics Technology Co., Ltd." is identified, and a desensitization rule is triggered by regular expression matching, setting... , If so, the desensitized output will be "Pr*******n", which preserves the readability of the context while preventing leakage; Record the entire operation log and perform data leakage prevention audits, and calculate the anomaly score of the operation behavior. : ,in, This is the current operation record. For the first Anomaly detection feature functions, These are the weight coefficients of the corresponding feature functions. The total number of anomaly detection features; when An anti-leakage alarm is triggered when the value exceeds the preset alarm threshold.

[0043] In this embodiment, three anomaly detection feature functions are configured: : The function for abnormal daily creative downloads is defined as follows The normalized score is triggered when the number of downloads exceeds 50 on a given day. : Non-working time access exception function, defined as ; Cross-departmental unauthorized access function, defined as .

[0044] Set the weights of each feature , , (Unauthorized access carries the highest risk and carries the greatest weight), preset alarm threshold. During an audit, it was detected that account 'o' downloaded 30 cross-departmental creative ideas at 1:30 AM (15 of which had inaccessible permissions). The calculation showed that: (Outside of working hours) (Unauthorized access) The system immediately triggers an anti-leakage alarm and performs the following actions: (1) automatically freezes the account's access permissions; (2) sends an alarm notification to the security administrator via Lark; (3) marks all operation logs during that period as pending audit. This ensures the security and traceability of patent idea transfer.

[0045] In summary, this embodiment achieves intelligent and closed-loop management of the entire process of patent ideas from submission, retrieval, evaluation to feedback by decoupling and coordinating multiple agents, combined with semantic parsing, hybrid retrieval, joint judgment model and dynamic priority scheduling. At the same time, it ensures the asset security of core data through multiple security mechanisms.

[0046] Second Embodiment Please see Figure 8 As shown, based on the same concept, this invention also provides a patent idea intelligent screening and evaluation system based on multi-agent collaboration, including: The creative semantic parsing module 201 is used by the creative intelligent agent to receive the patent creative text submitted by the user, perform semantic parsing on the patent creative text, and extract the technical application scenarios, pain points of existing solutions, creative descriptions and core innovations. The multi-database parallel retrieval module 202 is used by the retrieval agent to call the semantic retrieval engine based on the technical application scenario and core innovation points, and perform parallel retrieval in the patent database, academic literature database and technical report database to generate a retrieval result set; The multi-dimensional evaluation and reasoning module 203 is used by the evaluation agent to evaluate and reason about the patent idea from multiple dimensions such as novelty, practicality, technical effect, scenario difference and technical path coverage based on the search result set, and generate an evaluation conclusion. The evaluation conclusion collaborative execution module 204 is used by the collaborative intelligent agent to fill the evaluation conclusion back into the corresponding record in the preset management system and send the evaluation notification through the preset notification channel. The timed monitoring and cyclic scheduling module 205 is used by the collaborative intelligent agent to monitor the submission of new ideas according to the preset scheduling rules, and to trigger the cyclic execution of the idea semantic parsing module 201, the multi-database parallel retrieval module 202, the multi-dimensional evaluation reasoning module 203, and the evaluation conclusion collaborative execution module 204 when an idea to be evaluated is detected. The multi-agent communication and collaboration module 206 is used for message interaction, task orchestration and state synchronization between the creative agent, the retrieval agent, the evaluation agent and the collaborative agent through a preset multi-agent communication protocol, so as to realize the collaborative execution of patent idea screening and evaluation processing.

[0047] Third Embodiment Based on the same concept, the present invention also provides a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor causes the processor to perform the steps of a patent idea intelligent screening and evaluation method based on multi-agent collaboration as described in the embodiments.

[0048] Based on the same concept, the present invention also provides a storage medium storing computer-readable instructions, characterized in that, when the computer-readable instructions are executed by one or more processors, the one or more processors cause the one or more processors to perform the steps of a patent idea intelligent screening and evaluation method based on multi-agent collaboration as described in any one embodiment.

[0049] It is understood that, regarding the aforementioned method for intelligent screening and evaluation of patent ideas based on multi-agent collaboration, if all components are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer server or a network device, etc.) to execute all or part of the steps of the methods in the various embodiments of this invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0050] Computer-readable storage media may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable storage medium may also be any readable medium other than a readable storage medium that can transmit, propagate, or transfer a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0051] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

[0052] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for intelligent screening and evaluation of patent ideas based on multi-agent collaboration, characterized in that, Includes the following steps: S1: The creative intelligent agent receives the patent creative text submitted by the user, performs semantic analysis on the patent creative text, and extracts the technical application scenarios, pain points of existing solutions, creative descriptions and core innovations; S2: The retrieval agent calls the semantic retrieval engine based on the technology application scenario and the core innovation point to perform parallel retrieval in the patent database, academic literature database and technology report database, and generate a retrieval result set; S3: The evaluation agent evaluates and reasones the patent idea from multiple dimensions, including novelty, practicality, technical effect, scenario difference, and technical path coverage, based on the search result set, and generates an evaluation conclusion. S4: The collaborative intelligent agent fills the evaluation conclusion back into the corresponding record in the preset management system and sends the evaluation notification through the preset notification channel; S5: The collaborative intelligent agent monitors the submission of new ideas according to the preset scheduling rules, and triggers steps S1 to S4 to be executed in a loop when an idea to be evaluated is detected. S6: The creative agent, retrieval agent, evaluation agent, and collaborative agent interact with each other through a preset multi-agent communication protocol to arrange messages, schedule tasks, and synchronize states, so as to achieve collaborative execution of patent idea screening and evaluation processing.

2. The method for intelligent screening and evaluation of patent ideas based on multi-agent collaboration as described in claim 1, characterized in that, In step S1, semantic parsing of the patent idea text includes: The patent creative text is used to identify intent based on a pre-trained language model in order to determine the core appeal of the creative idea. Technical features are extracted from the patent creative text based on named entity recognition technology to obtain technical entity phrases. Based on semantic similarity, the innovative points of the technical entity phrases are clustered, and the unstructured creative text is transformed into structured technical application scenarios, pain points of existing solutions, creative descriptions and core innovative points; The semantic similarity is calculated in the following way: ,in, and These represent two technical entity phrases, and These represent the corresponding semantic vector representations; when When the similarity exceeds a preset threshold, the corresponding technical entity phrases will be grouped into the same innovation point cluster.

3. The method for intelligent screening and evaluation of patent ideas based on multi-agent collaboration as described in claim 1, characterized in that, In step S2, the semantic retrieval engine is invoked to perform multi-database parallel retrieval, including: A hybrid retrieval strategy is constructed based on the application scenarios of the technology and the core innovations, employing retrieval enhancement generation technology. The hybrid retrieval strategy combines vector retrieval and keyword retrieval, and performs semantic matching simultaneously in patent databases, academic literature databases, and technical report databases. The matching results returned by each database are aggregated and sorted based on relevance scores to generate a search result set containing highly relevant patents and documents. Among them, the relevance score Calculate using the following formula: ,in, The semantic similarity score for vector retrieval. This is the matching score for keyword retrieval. The weighting coefficient has a range of values. This is used to adjust the weight ratio of semantic matching and exact matching in the final ranking.

4. The method for intelligent screening and evaluation of patent ideas based on multi-agent collaboration according to claim 1, characterized in that, In step S3, the multi-dimensional evaluation reasoning includes: The block compares the core innovation points with the existing technical features in the search result set to evaluate the coverage of technical paths; The scenario difference between the patented idea and existing technologies is calculated, and a quantitative score is given by combining practicality indicators and technical effect parameters; wherein, the scenario difference... Calculate using the following formula: ,in, This represents the scene semantic vector corresponding to the patented idea. This represents the scene semantic vector corresponding to the existing technology; Based on the combined novelty evaluation results and quantitative scoring results, a comprehensive evaluation score is calculated, matched against preset evaluation rules, and an assessment conclusion including an evaluation level and explanatory reasons is generated; the comprehensive evaluation score Calculate using the following formula: ,in, The score represents the novelty factor. Indicates the practicality score. Indicates the score for technical effectiveness. Indicates the coverage of technology paths. to These are the weight coefficients for the corresponding dimensions, and they satisfy... .

5. The method for intelligent screening and evaluation of patent ideas based on multi-agent collaboration according to claim 4, characterized in that, The evaluation conclusions include: The evaluation conclusion is generated based on the joint judgment result of whether the core innovation points are covered by existing technologies and the degree of difference in scenarios, combined with the comprehensive evaluation score. The joint determination result is obtained by calculating a joint determination value. Confirmed, the calculation formula is: ,in, This is an adjustment coefficient, and its value range is... ; when If the score exceeds the preset coverage threshold, it is determined that the core innovation points are partially covered or highly related. If the comprehensive evaluation score is lower than the first preset threshold, an evaluation conclusion that the application is not recommended is generated. If the comprehensive evaluation score is between the first and second preset thresholds, an evaluation conclusion that the application should be adjusted and resubmitted is generated, and the differentiated details or technical adjustment directions that need to be clarified are specified in the explanatory reasons. when If the difference is less than the preset difference threshold and the comprehensive evaluation score is higher than the second preset threshold, it is determined that the core innovation point is not covered and the difference is significant. An evaluation conclusion recommending the application is generated, and the basis supporting the novelty of the core innovation point is output.

6. The method for intelligent screening and evaluation of patent ideas based on multi-agent collaboration according to claim 1, characterized in that, In step S4, the evaluation conclusion is populated back into the preset management system and a notification is pushed, including: The collaborative agent analyzes the evaluation conclusions and extracts the evaluation level, explanatory reasons, and associated original creative identifiers; Based on the original creative identifier, the evaluation level and explanatory reasons are automatically written into the designated field of the corresponding record in the management system to form a data closed loop of creative submission, evaluation processing and result retention; Simultaneously, a message push event is triggered, the target notification channel is determined according to system configuration or user preferences, the evaluation level and explanatory reasons are encapsulated into a message format adapted to the target notification channel, and the corresponding interface is called to complete the targeted message push; Among them, the push priority of the k-th notification channel Determine using the following formula: ,in, Indicates the historical success rate of reaching an agreement. Indicates timeliness score, Indicates user preference weights, , , Let be the weight coefficient, and satisfy... .

7. The method for intelligent screening and evaluation of patent ideas based on multi-agent collaboration according to claim 1, characterized in that, In step S5, automatically monitoring new creative submissions and triggering cyclic execution according to preset scheduling rules includes: The collaborative intelligent agent initiates a monitoring task according to a preset scheduling cycle, scanning the creative submission records in the management system whose status is marked as new, pending processing, or unevaluated. After a new creative idea is detected, the corresponding creative text is automatically extracted, and the creative agent, retrieval agent, evaluation agent and collaborative agent are triggered in sequence according to the trigger priority of the record to be processed to execute the processing flow. Among them, the triggering priority Calculate using the following formula: ,in, This indicates the duration of the record waiting time. Indicates the urgency of the business to which the creative idea belongs. Indicates the submitter's priority. , , These are the weighting coefficients.

8. The method for intelligent screening and evaluation of patent ideas based on multi-agent collaboration according to claim 1, characterized in that, This also includes responses to comments from intelligent agents assisting in the review process: The responding agent sends a supplementary search request to the searching agent through the multi-agent communication protocol; The search agent re-executes a multi-database search based on the technical disputes and returns the supplemented relevant comparative literature to the response agent. Based on the supplementary comparative literature and the original patent idea, the response agent generates a rebuttal strategy, key points of the response, and directions for modifying the application documents in response to examination inquiries, and pushes them to the intellectual property specialist. The technical dispute matching degree G is calculated according to the following formula: ,in, This indicates the number of feature terms in the supplementary comparative literature that match the technical dispute. This represents the total number of features of the technical dispute.

9. The method for intelligent screening and evaluation of patent ideas based on multi-agent collaboration according to claim 1, characterized in that, It also includes security control steps: When data is transmitted between the intelligent agents, the transmitted patent idea text and evaluation data are encrypted. Users who access and operate the management system shall be subject to hierarchical access verification to restrict unauthorized access. Sensitive information in the text is anonymized before data backfilling and notification push. The desensitization process uses a character mask desensitization function, defined as follows: ,in, For the text field to be processed, Indicates reserved fields The former The character and the following Each character, with the middle part replaced by a preset mask character; Record the entire operation log and perform data leakage prevention audits, and calculate the anomaly score of the operation behavior. : ,in, This is the current operation record. For the first Anomaly detection feature functions, These are the weight coefficients of the corresponding feature functions. The total number of anomaly detection features; when An anti-leakage alarm is triggered when the value exceeds the preset alarm threshold.

10. A patent idea intelligent screening and evaluation system based on multi-agent collaboration, characterized in that, include: The creative semantic parsing module is used by the creative intelligent agent to receive the patent creative text submitted by the user, perform semantic parsing on the patent creative text, and extract the technical application scenarios, pain points of existing solutions, creative descriptions and core innovations. The multi-database parallel retrieval module is used by the retrieval agent to call the semantic retrieval engine based on the technology application scenario and the core innovation point, and to perform parallel retrieval in the patent database, academic literature database and technology report database to generate a retrieval result set; The multi-dimensional evaluation and reasoning module is used by the evaluation agent to evaluate and reason about the patent idea from multiple dimensions, including novelty, practicality, technical effect, scenario difference, and technical path coverage, based on the search result set, and generate an evaluation conclusion. The evaluation conclusion collaborative execution module is used by the collaborative intelligent agent to backfill the evaluation conclusion into the corresponding record in the preset management system and send the evaluation notification through the preset notification channel. The timed monitoring and cyclic scheduling module is used by the collaborative intelligent agent to monitor the submission of new ideas according to preset scheduling rules, and to trigger the cyclic execution of the idea semantic parsing module, the multi-database parallel retrieval module, the multi-dimensional evaluation reasoning module and the evaluation conclusion collaborative execution module when an idea to be evaluated is detected. The multi-agent communication and collaboration module is used for message interaction, task orchestration, and state synchronization between the creative agent, retrieval agent, evaluation agent, and collaboration agent through a preset multi-agent communication protocol, so as to realize the collaborative execution of patent idea screening and evaluation processing.

Images