Service discovery and resource matching method and device of agent service system, and computer readable storage medium

Abstract

A service discovery and resource matching method and device of an agent service system, and a computer readable storage medium, the method comprising: creating an agent service domain name database based on a preset semantic description framework, the database recording resource record information of each agent; performing semantic understanding on a received agent service query request to obtain a standardized query text, and performing vectorization processing to obtain a semantic query vector; preliminarily screening the resource record information according to the standardized query text; sorting the agents according to the vector similarity between the semantic query vector and the resource record information of the agents in the preliminary selection range to obtain candidate agents; and screening the candidate agents according to the context of the agent service query request to obtain and feed back a matching result. The application can solve the application ecological gap caused by the coexistence of multiple protocols, improve the semantic universality of heterogeneous agent service resources, and support efficient interconnection of agent service resources.

Description

Technical Field

[0001] This invention relates to the field of intelligent agent service technology, and in particular to a service discovery and resource matching method and apparatus, and a computer-readable storage medium for an intelligent agent service system. Background Technology

[0002] With the large-scale application of new technologies such as artificial intelligence, large-scale models, and computing networks, agent services have become an important part of the digital economy. Accurately matching service demands and scientifically organizing and scheduling massive amounts of heterogeneous agent services have become significant challenges for their sustainable development. Regarding unified resource definition, the standardized identification, registration, and management of agent service resources have reached a point where each agent service can be assigned a unique and verifiable digital identity, and its lifecycle, permissions, and capabilities can be systematically managed. This forms the basis for agent discovery, mutual trust, collaboration, and oversight. Currently, research on agent service discovery, decision evaluation, and scheduling technologies has made some progress in system architecture and protocols, and has initially established a basic model for agent network service discovery and collaborative operation.

[0003] For example, some research has proposed intelligent agent protocol systems that clearly define relevant entities such as intelligent agent registration service providers and credential issuers, providing a standardized reference for the entire process of intelligent agents from registration and authentication to discovery. The Agent Union's ACP (Agent Communication Protocol) uses a unique AID (Intelligent Agent Identity) and an access point network to achieve intelligent agent addressing and communication (the AID is a second-level domain name obtained through wildcard domain name resolution of the access point; agents use the AID to find each other and communicate). Google's A2A (Agent-to-Agent Protocol) relies on "Agent Cards" (standardized information carriers used to carry core intelligent agent information) to describe identity and can rely on registry centers such as Nacos (Dynamic Naming and Configuration Service) to achieve service registration and discovery.

[0004] However, the aforementioned intelligent agent service discovery and resource matching solutions still face several problems, including insufficient semantic universality, low accuracy of matching and scheduling, and weak dynamics in service discovery. First, the technology system is fragmented, with multiple protocols such as MCP (Model Context Protocol), ANP (Agent Network Protocol), A2A, and ACP coexisting in the market. These protocols differ in their semantic identification methods and service architectures, lacking a unified and compatible semantic description standard. This makes it difficult for intelligent agents across protocols to interconnect at the identification and discovery levels, resulting in significant application ecosystem barriers.

[0005] Secondly, existing traditional matching mechanisms mostly rely on service names, function types, or single requirements for retrieval and allocation, making it difficult to achieve refined, granular, and real-time resource allocation and scheduling management. This makes it difficult to meet users' multi-dimensional and accurate matching needs for intelligent agent services, resulting in low matching and scheduling accuracy. Furthermore, heterogeneous resource identifiers and corresponding information management models lack dynamic adaptability. Single identity management systems (such as OAuth, Open Authorization) and static permission models (such as RBAC, Role-Based Access Control) cannot grasp the dynamic global resource situation of the network, making it difficult to adapt to the characteristics of variable agent tasks and short lifecycles, and hindering dynamic resource allocation and scheduling management. How to dynamically evaluate, track, and supervise the service status of intelligent agents is also a key issue that requires attention. Summary of the Invention

[0006] One of the objectives of this invention is to provide a service discovery and resource matching method for an intelligent agent service system, which solves the application ecosystem barriers caused by the coexistence of multiple protocols in intelligent agents, improves the semantic universality and interoperability of heterogeneous intelligent agent service resources, and supports efficient interoperability and collaborative application of cross-domain, cross-network, and multi-type intelligent agent service resources.

[0007] In a first aspect, embodiments of the present invention provide a service discovery and resource matching method for an intelligent agent service system, comprising: Based on a preset semantic description framework, an intelligent agent service domain name database is created, which records the resource record information of each intelligent agent. The received intelligent agent service query request is semantically understood to obtain a standardized query text, and the standardized query text is vectorized to obtain a semantic query vector. Based on the standardized query text, the resource record information in the intelligent agent service domain name database is initially screened to obtain intelligent agents included in the initial selection range; Based on the similarity between the semantic query vector and the resource record information of the agents included in the initial selection range, the agents included in the initial selection range are sorted, and candidate agents are obtained based on the sorting results. Based on the context of the agent service query request, the candidate agents are filtered, and the matching results are obtained and returned.

[0008] In one possible embodiment, the semantic description framework includes at least one of the following: Service metadata of an agent is used to record one or more of the agent's globally unique identifier, name and version number, provider information, registration time and lifespan, and identity authentication tags. The functional description data of the agent is used to record one or more of the agent's functional domain labels, natural language descriptions of capabilities, and core functional labels. Performance description data of an agent, used to record one or more of the agent's performance benchmarks, service quality attributes, and resource requirements; Security description data of an intelligent agent is used to record one or more of the intelligent agent's service access endpoints, authentication protocols, access control policies, and input and output modes.

[0009] In one possible embodiment, the received agent service query request is semantically understood to obtain a standardized query text, and the standardized query text is then vectorized to obtain a semantic query vector, including: The received intelligent agent service query request is processed using a large language model to understand the query requirements. Then, according to the data distribution in the semantic description framework, the query requirements are transformed into the standardized query text; The standardized query text is vectorized using a semantic vector processing model to obtain a semantic query vector.

[0010] In one possible embodiment, preliminary screening of resource record information in the agent service domain name database based on the standardized query text includes: Select one or more keywords from the standardized query text, search the intelligent agent service domain name database, and match the keywords with the resource record information; The overall keyword matching degree of each agent is determined based on the matching degree with one or more keywords, and agents with an overall keyword matching degree higher than a preset threshold are included in the initial selection range.

[0011] In one possible embodiment, ranking the agents in the initial selection range based on the vector similarity between the semantic query vector and the resource record information of the agents included in the initial selection range includes: The resource record information of the intelligent agents included in the initial selection scope is subjected to the vectorization process to obtain semantic service vectors; Calculate the vector similarity between the semantic query vector and the semantic service vector, and sort them according to the vector similarity. A preset number of agents are selected based on the sorting results to obtain the candidate agents.

[0012] In one possible embodiment, filtering the candidate agents based on the context of the agent service query request includes: Based on the context of the agent service query request, a comprehensive score is calculated for each candidate agent, the comprehensive score including a weighted sum of factor values; The factor values ​​include one or more of the following: semantic similarity factor, service quality factor, performance constraint factor, and system optimization factor. The semantic similarity factor includes the vector similarity, the service quality factor includes dynamic evaluation data of agent service quality, the performance constraint factor includes mandatory performance requirements, and the system optimization factor includes one or more of load balancing, service deployment location, network latency, and cost. The candidate agents are then sorted and filtered based on the comprehensive score.

[0013] In one possible embodiment, it further includes: After the agent responds to the agent service query request, it evaluates and verifies the completed task status, and dynamically adjusts one or more of the semantic description framework, preliminary screening process, sorting process and context screening process based on the evaluation and verification results. The evaluation and verification include one or more of the following: functional completion, performance achievement, and system reliability. The functional completion rate is used to verify whether the agent accurately completes its function. The performance achievement is used to test the response latency, throughput, and resource consumption of the agent. The system reliability is used to evaluate the system's operational capability during the process of the agent completing the task.

[0014] Secondly, embodiments of the present invention provide a service discovery and resource matching apparatus for an intelligent agent service system, used to implement the service discovery and resource matching method for an intelligent agent service system as described above, including: The storage module is used to create an intelligent agent service domain name database based on a preset semantic description framework. The intelligent agent service domain name database records the resource record information of each intelligent agent. The parsing module is used to perform semantic understanding on the received intelligent agent service query request, obtain standardized query text, and perform vectorization processing on the standardized query text to obtain semantic query vector; The preliminary screening module is used to perform preliminary screening of resource record information in the intelligent agent service domain name database based on the standardized query text, so as to obtain intelligent agents included in the preliminary selection range; The sorting module sorts the agents included in the initial selection range according to the vector similarity between the semantic query vector and the resource record information of the agents included in the initial selection range, and obtains candidate agents based on the sorting results. The context filtering module filters the candidate agents based on the context of the agent service query request, and obtains and returns the matching results.

[0015] Thirdly, embodiments of the present invention also provide a computer-readable storage medium, which is a non-volatile storage medium or a non-transient storage medium, on which a computer program is stored, and the computer program is executed by a processor to perform the steps of any of the methods described above.

[0016] Fourthly, embodiments of the present invention also provide another electronic device, including a memory and a processor, wherein the memory stores a computer program executable on the processor, and the processor executes the steps of any of the methods described above when running the computer program.

[0017] Fifthly, embodiments of the present invention also provide a computer program product, characterized in that it includes a computer program / instruction, which, when executed by a processor, implements the steps described above.

[0018] Compared with the prior art, the technical solution of the embodiments of the present invention has the following advantages: The solution of this invention constructs an intelligent agent service domain name database based on a semantic description framework. Combined with the hierarchical structure characteristics of domain name system registration and resource management, it can effectively solve the application ecosystem barriers caused by the coexistence of multiple protocols of intelligent agents, improve the semantic universality and interoperability of heterogeneous intelligent agent service resources, and support the efficient interoperability and collaborative application of cross-domain, cross-network, and multi-type intelligent agent service resources. Furthermore, based on standardized query text, a rapid coarse screening is performed, followed by preliminary sorting using vector similarity, and finally, a fine sorting is performed using dynamic factor values. This achieves accurate and efficient service filtering and sorting, enabling refined, granular, and real-time resource allocation and scheduling management. It meets users' multi-dimensional and accurate matching needs for intelligent agent services, significantly improving matching and scheduling accuracy and resource utilization. Furthermore, based on the agent service evaluation and feedback learning mechanism, the semantic description framework and various parameters in the matching process can be dynamically adjusted to adapt to the characteristics of agent tasks being varied and having short lifecycles. This enables dynamic and real-time resource allocation and scheduling management from a global perspective, allowing for dynamic evaluation, tracking, and supervision of agent service status, thereby improving the overall stability and security of the system. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a service discovery and resource matching method of an intelligent agent service system in one embodiment of this application; Figure 2 This is a schematic diagram of a service discovery and resource matching device for an intelligent agent service system according to one embodiment of this application. Detailed Implementation

[0020] To make the above-mentioned objectives, features and beneficial effects of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] In one embodiment of the present invention, reference is made to... Figure 1 This invention provides a service discovery and resource matching method for an intelligent agent service system. This method can be applied to an intelligent agent service management system, which provides users with intelligent agent discovery and matching services. Specifically, the method may include: Step 101: Based on the preset semantic description framework, create an agent service domain name database. The agent service domain name database records the resource record information of each agent. In one embodiment of the present invention, an agent refers to an intelligent program or service unit that possesses perception, understanding, and execution capabilities and can autonomously complete tasks; agent services emphasize the functions and service capabilities provided by the agent; agent service resources refer to the sum of various resource information that support the operation of agent services and can be discovered and allocated within the system. The agent service system, based on an innovatively designed hierarchical semantic analysis model, combines the domain name system and related technologies to conduct agent service resource discovery and scheduling management services. The hierarchical semantic analysis model includes the overall model used by the agent service system in each embodiment of this application for service discovery and resource matching.

[0022] The traditional Domain Name System (DNS), as a crucial internet infrastructure, is the most important form of internet resource addressing and resolution, exhibiting significant stability and reliability. The Intelligent Agent Service System builds an Intelligent Agent Service Domain Name Database based on the DNS, unifying the registration information and dynamic management of Intelligent Agent Service resources, and incorporating a hierarchical data management and resolution service node design architecture. The Intelligent Agent Service Domain Name Database provides domain name resolution (DNS resolution) and management for Intelligent Agents. Domain name resolution includes translating the service name or domain name of the Intelligent Agent into a network address; management includes adding, deleting, modifying, and querying Intelligent Agents; and hierarchical management involves different levels of service nodes undertaking global data management, regional data management, and local resolution services, respectively, achieving distributed resolution and hierarchical data maintenance of Intelligent Agent Service Domain Names. Resource Record (RR) information is the core data unit stored in the DNS, used to describe the mapping relationship between a domain name (or Intelligent Agent) and its corresponding network information and service attributes. This invention fully utilizes internet service capabilities to provide cross-network, cross-business, and cross-system Intelligent Agent Service resource registration, resolution, allocation, and evaluation services, thereby supporting the unified management, dynamic resource matching, and service scheduling needs of heterogeneous Intelligent Agent Service resources.

[0023] The semantic description framework includes at least one of the following: service metadata of the agent, used to record one or more of the agent's globally unique identifier, name and version number, provider information, registration time and lifespan, and identity authentication tags; functional description data of the agent, used to record one or more of the agent's functional domain tags, capability natural language descriptions, and core function tags; performance description data of the agent, used to record one or more of the agent's performance benchmarks, service quality attributes, and resource requirements; and security description data of the agent, used to record one or more of the agent's service access endpoints, authentication protocols, access control policies, and input and output modes.

[0024] Specifically, step 101 includes: defining a unified semantic description framework for intelligent agent services. Intelligent agent registration information, according to the requirements of this semantic description framework, is stored in the intelligent agent service domain name database in a specific format such as JSON, TXT, or YAML. The semantic description framework mainly includes basic data (or service metadata, including globally unique identifiers (UUIDs), names, etc.) describing the intelligent agent, semantic attributes (or functional description data, including functional domain tags, core functional tags, etc.), performance metrics (or performance description data, including accuracy, average response time, resource requirements, etc.), and security data (or security description data, including authentication protocols, interface specification frameworks, network addresses, etc.). Simultaneously, this model also allows users to flexibly add custom fields to adapt to the specific needs of different domains, while maintaining machine readability and a certain degree of semantic richness. Furthermore, the interface and performance fields included in the description can also be directly used as input metrics for subsequent evaluation modules. The hierarchical semantic analysis model includes this semantic description framework.

[0025] The semantic description framework for intelligent agent services can adopt formats such as JSON, TXT, or YAML. An embodiment of this invention illustrates partial data from an example. Specific information fields and content format requirements include, but are not limited to, the following pseudocode and Tables 1 to 4 below.

[0026] {

[0027] "asdp_version": "1.0.0", "metadata": { "agent_id": "urn:agent:patent-eagle-eye:v2.1", "name": "Industry Data Eagle Eye Analysis Intelligent Agent", "version": "2.1.0", "provider": { "name": "XXX Technology Co., Ltd". }, "registration_time": "2023-10-27T08:30:00Z", "last_update_time": "2023-11-05T14:20:00Z", "ttl": 3600, "status": "active" } "input_output_schema": { "api_endpoint": " / v1 / analyze / risk", "request_content_type": "application / json", "request_schema": { "$ref": "# / functional_description / core_functions / 0 / input_format" }, } "non_functional_description": { "performance_metrics": { "benchmark": { "p99_latency_ms": 1200, "throughput_rpm": 150, "accuracy": 0.92 } } Table 1. Service Metadata Y

[0028] Table 2. Functional Description Data G

[0029] Table 3. Performance Description Data X

[0030] Table 4. Security Description Data S

[0031] In this embodiment of the invention, by constructing an intelligent agent service domain name database based on a semantic description framework, and combining the registration and resource management hierarchical structure characteristics of the domain name system, metadata such as globally unique identifiers can be registered to achieve standardized descriptions of intelligent agent service functions and performance. At the same time, corresponding authentication protocols and network addresses are associated to complete resolution and access respectively, effectively solving the application ecosystem barriers caused by the coexistence of multiple intelligent agent protocols, improving the semantic universality and interoperability of heterogeneous intelligent agent service resources, and supporting efficient interoperability and collaborative application of cross-domain, cross-network, and multi-type intelligent agent service resources.

[0032] Step 102: Perform semantic understanding on the received intelligent agent service query request to obtain standardized query text, and perform vectorization processing on the standardized query text to obtain semantic query vector; In one embodiment of the present invention, step 102 includes: understanding the needs of the received intelligent agent service query request through a large language model to obtain the query needs; then converting the query needs into standardized query text according to the data distribution in the semantic description framework; and using a semantic vector processing model to vectorize the standardized query text to obtain a semantic query vector.

[0033] Specifically, step 102 also includes: the user sends an intelligent agent service query request through a DNS query request message on the client, and the intelligent agent service system understands the user's needs, which can be in the form of natural language description or code.

[0034] For user requirements in natural language description form, the input requirements are first obtained using a large language model parsing tool, and then processed into a structured semantic graph and standardized query text based on a semantic description framework. For user requirements in code form, they are processed into standardized query text based on a semantic description framework.

[0035] The standardized query text is then vectorized using a semantic vector processing model (such as BGE-M3 or text2vec) to obtain the semantic query vector Q. BGE-M3 (Bidirectional General Embedding Model-M3) is a multilingual embedding model; "M3" represents its three core characteristics: multi-functionality, multi-linguality, and multi-granularity. text2vec is an open-source library for text vectorization and semantic matching. The hierarchical semantic analysis model includes the large language model and semantic vector processing model used in this application embodiment, along with their corresponding parameters. The model used for processing can be dynamically optimized by incorporating dynamic data from the query request-service matching process. This invention embodiment includes, but is not limited to, the following pseudocode.

[0036] # 1. Structured text processing to obtain standardized query text: query_text = "[Intent] Analyze text security risk assessment [Constraints] Maximum latency 500ms ..." # 2. Encode a single text and obtain a dense embedding. embeddings = model.encode(query_text, return_dense=True, return_sparse=False, return_colbert_vecs=False) dense_vector = embeddings['dense_vecs'] # A NumPy array of shape [1, 1024] # The above dense_vector is a single semantic query vector Q, and dense_vecs are the core function tags in the function description data above. # 3. Batch encoding service description, building a vector library service_descriptions = ["Description of service 1...", "Description of service 2..."] service_embeddings = model.encode(service_descriptions, return_dense=True, batch_size=32)['dense_vecs'] This invention, based on the deep semantic understanding and requirement parsing capabilities of a semantic big model, processes user-input natural language or programmatic query requirements sequentially into a structured semantic graph, standardized query text, and vectors, and then enhances semantics using a semantic description framework. This addresses the problem that traditional keyword matching cannot understand the deep semantics and contextual relationships of queries.

[0037] Step 103: Perform preliminary screening of resource record information in the agent service domain name database based on standardized query text to obtain agents included in the initial selection range; In one embodiment of the present invention, step 103 includes: selecting one or more keywords in the standardized query text, searching in the agent service domain name database, and matching the keywords with resource record information; determining the comprehensive keyword matching degree of each agent based on the matching degree with one or more keywords, and including agents with a comprehensive keyword matching degree higher than a preset threshold in the initial selection range.

[0038] Specifically, step 103 also includes: in order to make a preliminary selection from a large number of intelligent agents, the intelligent agent service system adopts a fast matching method based on text keywords. For example, keywords in the standardized query text are selected and searched in the intelligent agent service domain name database according to the following formula (1). All intelligent agents with a calculated score greater than 0.8 (N>0.8) will be included in the preliminary selection range.

[0039] N=a + b (1)

[0040] Where N is the overall keyword matching degree, with a preset threshold of 0.8; Qg is the matching degree of each function description data (Table 2); a is the coefficient of the matching degree of the function description data, which can usually be set to 0.8; Qx is the matching degree of each performance description data (Table 3); b is the coefficient of the matching degree of the performance description data, which can usually be set to 0.2. This represents the sum of the matching degrees of the first three items in the functional description data (Table 2); This represents the sum of the matching degrees of the first three items in the performance description data (Table 2).

[0041] The hierarchical semantic analysis model includes formula (1) and its corresponding coefficients or thresholds in the embodiments of this application.

[0042] Step 104: Sort the agents included in the initial selection range according to the vector similarity between the semantic query vector and the resource record information of the agents included in the initial selection range, and obtain candidate agents according to the sorting results. In one embodiment of the present invention, step 104 includes: vectorizing the resource record information of the agents included in the preliminary selection scope to obtain semantic service vectors; calculating the vector similarity between the semantic query vector and the semantic service vector, and sorting them according to the vector similarity; selecting a preset number of agents according to the sorting results to obtain candidate agents.

[0043] Specifically, step 104 further includes: the intelligent agent service system requests to query the resource record information of the intelligent agents included in the initial selection scope, and converts it into a semantic service vector S according to the same semantic vector processing model (such as BGE-M3) as the semantic query vector Q. An efficient approximate nearest neighbor search index is constructed using the HNSW (Hierarchical Navigable Small World) algorithm. Further refined matching is performed through similarity calculation between the semantic query vector Q and the semantic service vector S, sorted in descending order, and a list of the top 10 candidate services based on vector similarity is returned. Here, vector similarity is the cosine similarity between the semantic query vector Q and the semantic service vector S, and 10 is a preset number. Embodiments of this invention include, but are not limited to, the following pseudocode.

[0044] def find_similar_services(query_vector, top_k=10, index):

[0045] """

[0046] Find the most similar agent service for semantic query vector Q

[0047] query_vector: The BGE-M3 encoded vector Q queried by the user.

[0048] top_k: The number of most similar services returned.

[0049] """

[0050] # 1. Quick search in the HNSW index

[0051] neighbor_ids = index.search(query_vector, k=top_k, ef_search=80)

[0052] # 2. Obtain detailed service information

[0053] results = []

[0054] for service_id in neighbor_ids:

[0055] service_info = database.get_service_by_id(service_id)

[0056] # Calculate exact similarity (cosine similarity)

[0057] service_vector = get_vector_by_id(service_id)

[0058] similarity = cosine_similarity(query_vector, service_vector)

[0059] results.append({

[0060] 'service_id': service_id, 'similarity': similarity, 'metadata': service_info }) # 3. Sort by similarity and return results.sort(key=lambda x: x['similarity'], reverse=True) return results[:top_10] Step 105: Based on the context of the agent service query request, filter the candidate agents and obtain and return the matching results.

[0061] In one embodiment of the present invention, step 105 includes: calculating a comprehensive score for each candidate agent based on the context of the agent service query request, wherein the comprehensive score includes a weighted sum of factor values; wherein the factor values ​​include one or more of semantic similarity factors, service quality factors, performance constraint factors, and system optimization factors; the semantic similarity factor includes vector similarity, the service quality factor includes dynamic evaluation data of agent service quality, the performance constraint factor includes mandatory performance requirements, and the system optimization factor includes one or more of load balancing, service deployment location, network latency, and cost; and updating the ranking based on the comprehensive score to screen the candidate agents.

[0062] Specifically, step 105 also includes: calculating a comprehensive score for each candidate agent service according to the following formula (2); Score = (2) Here, Score represents the overall score, Fi represents the factor value, and Wi represents the weight of the factor value. The factor values ​​include semantic similarity factor W1 (which includes vector similarity, i.e., the cosine similarity between the semantic query vector Q and the semantic service vector S), service quality factor W2 (dynamic evaluation data of agent service quality, which may involve the content related to service quality attributes in Table 3), performance constraint factor W3 (specific mandatory performance requirements, which can lead to direct elimination if not met, and may involve the content related to performance benchmarks in Table 3), and system optimization factor W4 (selectively considering load balancing, service deployment location, network latency, cost, etc., and may involve the content related to resource requirements in Table 3 and the relevant content in Table 4). The above factor values ​​and corresponding weights can be adjusted according to the context of user needs.

[0063] Finally, based on the comprehensive score calculated according to formula (2), a ranked list of recommended services is obtained, enabling further screening of candidate agents. The agent service system returns the list of recommended services and the corresponding resource summary information to the user client for selection. The hierarchical semantic analysis model includes formula (2) and its corresponding factor values ​​and weights in the embodiments of this application.

[0064] The multi-dimensional and context-aware matching process supported by this invention includes steps 103-105: first, a rapid coarse screening is performed based on the structural semantic graph and standardized query text; then, a preliminary ranking is performed based on vector similarity; and finally, a fine ranking is performed based on dynamic factors such as service performance and load, achieving accurate and efficient service filtering and ranking. This algorithm can balance the weights of semantic relevance with comprehensive influencing factors such as service performance and quality, providing dynamic calculation and allocation services. It solves the problem that a single matching dimension (such as text similarity) cannot meet the comprehensive consideration of service quality and real-time status. It can meet users' multi-dimensional accurate matching needs for agent services, effectively improve the accuracy of matching scheduling, and flexibly meet the diverse needs of agent service content.

[0065] Step 106, the service discovery and resource matching method of the intelligent agent service system in one embodiment of this application further includes: after the intelligent agent completes the intelligent agent service query request, evaluating and verifying the completed task status, and dynamically adjusting one or more of the semantic description framework, preliminary screening process, ranking process, and context screening process according to the evaluation and verification results; wherein, the matching process includes the preliminary screening process, ranking process, and context screening process, the preliminary screening process, ranking process, and context screening process correspond to steps 103-105 respectively, and the evaluation and verification content includes one or more of the following: functional completion degree, performance realization degree, and system reliability; functional completion degree is used to verify whether the intelligent agent accurately completes its function; performance realization degree is used to test one or more of the intelligent agent's response latency, throughput, and resource consumption; system reliability is used to evaluate the system operation capability of the intelligent agent in the process of completing the task.

[0066] In one embodiment of the present invention, dynamically adjusting one or more of the semantic description framework, preliminary screening process, ranking process, and contextual screening process includes continuously optimizing the semantic description framework of the intelligent agent service system and dynamically adjusting the parameters of each model during the matching process, such as dynamically adjusting the parameters in the hierarchical semantic analysis model. Evaluating and verifying the completion status and performance indicators of the completed tasks can improve the accuracy of subsequent semantic matching of intelligent agent services. Specific evaluation and verification content may include: Success rate is used to confirm that the interface of the matched agent is successfully invoked; functional completion rate is used to verify whether the agent accurately completes its claimed functional tasks (such as recall and precision). The parameters of each model are dynamically adjusted during the matching process. For example, when users in a certain industry repeatedly select agents with high functional completion rates, the keyword matching coefficient related to functional completion rate can be increased in step 103. Performance achievement rate is used to test its response latency, throughput, resource consumption, etc. System reliability is used to evaluate the system optimization factor W4 (load balancing, security, network stability, etc.). The evaluation process can be carried out by combining automated benchmark testing, automatic evaluation based on benchmark models, and necessary manual review methods.

[0067] This invention presents a dynamic, configurable, and scalable intelligent agent service evaluation and feedback learning mechanism. Based on real-time matching operation monitoring and feedback (such as call success rate, functional completion, performance achievement, system reliability, and user ratings), it dynamically adjusts the semantic description framework and the weights and specific parameter values ​​of various evaluation indicators during the matching process. Through a configurable weight mechanism, it meets the intelligent agent service matching needs of different industry applications and business requirements. It can also grasp the global dynamic resource situation of the network, adapting to the characteristics of variable intelligent agent tasks and short lifecycles, achieving dynamic resource allocation and scheduling management, dynamically evaluating, tracking, and monitoring the intelligent agent service status, and flexibly meeting the dynamic management needs of intelligent agent service status.

[0068] Reference Figure 2 One embodiment of the present invention provides a service discovery and resource matching device 20 for an intelligent agent service system, which can be used to implement the service discovery and resource matching method of the intelligent agent service system as described above, including: The storage module 201 is used to create an agent service domain name database based on a preset semantic description framework. The agent service domain name database records the resource record information of each agent. The parsing module 202 is used to perform semantic understanding on the received agent service query request, obtain standardized query text, and perform vectorization processing on the standardized query text to obtain semantic query vector. The preliminary screening module 203 is used to perform preliminary screening on the resource record information in the agent service domain name database according to the standardized query text to obtain agents included in the preliminary selection range. The ranking module 204 ranks the agents included in the preliminary selection range according to the vector similarity between the semantic query vector and the resource record information of the agents included in the preliminary selection range, and obtains candidate agents according to the ranking results. The context screening module 205 filters the candidate agents according to the context of the agent service query request, obtains and returns the matching results.

[0069] In one embodiment of the present invention, the storage module 201, the parsing module 202, the preliminary filtering module 203, the sorting module 204, and the context filtering module 205 are connected in sequence.

[0070] In specific implementation, the execution process of the above-mentioned storage module 201, parsing module 202, preliminary filtering module 203, sorting module 204, and context filtering module 205 can be referred to steps 101 to 105, which will not be elaborated here.

[0071] One embodiment of the present invention also provides a computer-readable storage medium, which is a non-volatile storage medium or a non-transient storage medium, on which a computer program is stored, and the computer program is executed by a processor to perform the steps of any of the methods described above.

[0072] An embodiment of the present invention also provides another electronic device, including a memory and a processor, wherein the memory stores a computer program executable on the processor, and the processor executes the steps of any of the methods described above when running the computer program.

[0073] One embodiment of the present invention also provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the method described above.

[0074] In some embodiments, each module / unit can also communicate with Internet Protocol (IP) networks, such as the Internet, private IP networks, or other data networks.

[0075] In specific implementation, the modules / units included in the various devices and products described in the above embodiments can be software modules / units, hardware modules / units, or a combination of both.

[0076] For example, for various devices and products applied to or integrated into a chip, each module / unit can be implemented using hardware methods such as circuits, or at least some modules / units can be implemented using software programs that run on a processor integrated within the chip, while the remaining (if any) modules / units can be implemented using hardware methods such as circuits; for various devices and products applied to or integrated into a chip module, each module / unit can be implemented using hardware methods such as circuits, and different modules / units can be located in the same component (e.g., chip, circuit module, etc.) or different components of the chip module, or at least some modules / units can be implemented using hardware methods such as circuits. The components can be implemented using software programs that run on the processor integrated within the chip module. The remaining (if any) modules / units can be implemented using hardware methods such as circuits. For various devices and products applied to or integrated into the terminal, each of its components / units can be implemented using hardware methods such as circuits. Different modules / units can be located in the same component (e.g., chip, circuit module, etc.) or in different components within the terminal. Alternatively, at least some modules / units can be implemented using software programs that run on the processor integrated within the terminal, while the remaining (if any) modules / units can be implemented using hardware methods such as circuits.

[0077] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include ROM, RAM, disk, or optical disk, etc.

[0078] While the present invention has been disclosed above, it is not limited thereto. Any person skilled in the art can make various modifications and alterations without departing from the spirit and scope of the invention; therefore, the scope of protection of the present invention should be determined by the scope defined in the claims.

Claims

1. A service discovery and resource matching method for an intelligent agent service system, characterized in that, include: Based on a preset semantic description framework, an intelligent agent service domain name database is created, which records the resource record information of each intelligent agent. The received intelligent agent service query request is semantically understood to obtain a standardized query text, and the standardized query text is vectorized to obtain a semantic query vector. Based on the standardized query text, the resource record information in the intelligent agent service domain name database is initially screened to obtain intelligent agents included in the initial selection range; Based on the similarity between the semantic query vector and the resource record information of the agents included in the initial selection range, the agents included in the initial selection range are sorted, and candidate agents are obtained based on the sorting results. Based on the context of the agent service query request, the candidate agents are filtered, and the matching results are obtained and returned.

2. The service discovery and resource matching method of the intelligent agent service system as described in claim 1, characterized in that, The semantic description framework includes at least one of the following: The service metadata of the intelligent agent is used to record one or more of the following: the intelligent agent's globally unique identifier, name and version number, provider information, registration time and survival time, and identity authentication tag. The functional description data of the intelligent agent is used to record one or more of the intelligent agent's functional domain labels, natural language descriptions of capabilities, and core functional labels; The performance description data of the intelligent agent is used to record one or more of the intelligent agent's performance benchmark, service quality, and resource requirements. Security description data of the intelligent agent is used to record one or more of the intelligent agent's service access endpoints, authentication protocols, access control policies, and input and output modes.

3. The service discovery and resource matching method of the intelligent agent service system as described in claim 1, characterized in that, The received intelligent agent service query request is semantically understood to obtain a standardized query text, and the standardized query text is then vectorized to obtain a semantic query vector, including: The received intelligent agent service query request is processed using a large language model to understand the query requirements. Then, according to the data distribution in the semantic description framework, the query requirement is transformed into the standardized query text; The standardized query text is vectorized using a semantic vector processing model to obtain a semantic query vector.

4. The service discovery and resource matching method of the intelligent agent service system as described in claim 1, characterized in that, The preliminary screening of resource record information in the intelligent agent service domain name database based on the standardized query text includes: Select one or more keywords from the standardized query text, search the intelligent agent service domain name database, and match the keywords with the resource record information; The overall keyword matching degree of each agent is determined based on the matching degree with one or more keywords, and agents with an overall keyword matching degree higher than a preset threshold are included in the initial selection range.

5. The service discovery and resource matching method of the intelligent agent service system as described in claim 1, characterized in that, The ranking of the agents in the initial selection range based on the similarity between the semantic query vector and the resource record information of the agents included in the initial selection range includes: The resource record information of the intelligent agents included in the initial selection scope is vectorized to obtain semantic service vectors; Calculate the vector similarity between the semantic query vector and the semantic service vector, and sort them according to the vector similarity. A preset number of agents are selected based on the sorting results to obtain the candidate agents.

6. The service discovery and resource matching method of the intelligent agent service system as described in claim 1, characterized in that, The filtering of candidate agents based on the context of the agent service query request includes: Based on the context of the agent service query request, a comprehensive score is calculated for each candidate agent, the comprehensive score including a weighted sum of factor values; The factor values ​​include one or more of the following: semantic similarity factor, service quality factor, performance constraint factor, and system optimization factor. The semantic similarity factor includes the vector similarity, the service quality factor includes dynamic evaluation data of the agent's service quality, the performance constraint factor includes mandatory performance requirements, and the system optimization factor includes one or more of the following: load balancing, service deployment location, network latency, and cost. The candidate agents are then sorted and filtered based on the comprehensive score.

7. The service discovery and resource matching method of the intelligent agent service system as described in claim 1, characterized in that, Also includes: After the agent responds to the agent service query request, it evaluates and verifies the completed task status, and dynamically adjusts one or more of the semantic description framework, preliminary screening process, sorting process and context screening process based on the evaluation and verification results. The evaluation and verification include one or more of the following: functional completion, performance achievement, and system reliability. The functional completion rate is used to verify whether the agent accurately completes its function. The performance achievement is used to test one or more of the agent's response latency, throughput, and resource consumption. The system reliability is used to evaluate the system's operational capability during the process of the agent completing the task.

8. A service discovery and resource matching apparatus for an intelligent agent service system, used to implement the service discovery and resource matching method for an intelligent agent service system as described in any one of claims 1 to 7, characterized in that, include: The storage module is used to create an intelligent agent service domain name database based on a preset semantic description framework. The intelligent agent service domain name database records the resource record information of each intelligent agent. The parsing module is used to perform semantic understanding on the received intelligent agent service query request, obtain standardized query text, and perform vectorization processing on the standardized query text to obtain semantic query vector; The preliminary screening module is used to perform preliminary screening of resource record information in the intelligent agent service domain name database based on the standardized query text, so as to obtain intelligent agents included in the preliminary selection range; The sorting module sorts the agents included in the initial selection range according to the vector similarity between the semantic query vector and the resource record information of the agents included in the initial selection range, and obtains candidate agents based on the sorting results. The context filtering module filters the candidate agents based on the context of the agent service query request, and obtains and returns the matching results.

9. A computer-readable storage medium, wherein the computer-readable storage medium is a non-volatile storage medium or a non-transient storage medium, and a computer program is stored thereon, characterized in that, The computer program is executed by the processor to perform the steps of the method according to any one of claims 1 to 7.

10. An electronic device comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor runs the computer program, it performs the steps of the method described in any one of claims 1 to 7.

11. A computer program product, characterized in that, It includes a computer program / instruction, which, when executed by a processor, implements the steps of the method as described in any one of claims 1 to 7.

Images