Multi-agent cooperative interaction method and device based on capability matching, electronic equipment and computer readable storage medium

Abstract

This application discloses a multi-agent collaborative interaction method, device, electronic device, and computer-readable storage medium based on capability matching. The method includes: receiving a collaboration request published by a first agent; determining the capability fit, capability gap, and credibility of each candidate agent based on the collaboration request and the capability information of each candidate agent; determining the comprehensive matching score of each candidate agent according to the above three dimensions; selecting a second agent that matches the collaboration request based on the comprehensive matching score and pushing the information of the second agent to the first agent; and establishing a collaboration relationship between the first agent and the second agent in response to the first agent initiating a collaboration invitation to the second agent, and executing the collaboration task corresponding to the collaboration request based on the collaboration relationship. This achieves fully automated interaction from request matching to collaboration completion in a multi-agent system, improving the accuracy of collaboration matching and task execution efficiency.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, apparatus, electronic device, and computer-readable storage medium for multi-agent collaborative interaction based on capability matching. Background Technology

[0002] With the rapid development of artificial intelligence technology, a single intelligent agent can only cover capabilities in a single sub-domain, exhibiting significant functional limitations. For example, an agent skilled in data crawling lacks natural language parsing capabilities, and an agent specializing in image recognition cannot perform time-series data prediction. Faced with complex and comprehensive tasks, it is difficult to complete them independently relying on a single intelligent agent; therefore, multi-agent collaboration has become the core approach to solving complex business scenarios.

[0003] Currently, the mainstream technologies used for intelligent agent collaboration fall into four categories, each with its own limitations: centralized scheduling systems rely on central order dispatch, and intelligent agents lack autonomous matching capabilities; computing power trading platforms only trade hardware resources and do not support algorithmic matching; human social platforms are adapted to natural persons and cannot meet the needs of intelligent agents for identity, structured data, and automatic matching in automated collaboration; traditional intelligent agent yellow pages can only perform simple keyword searches, lack experience quantification and trust transfer mechanisms, and have low matching accuracy.

[0004] In summary, existing multi-agent collaboration support solutions suffer from various shortcomings, including rigid scheduling modes, limited support for hardware resource transactions, inadequate adaptation to human interaction, and rudimentary matching mechanisms. These shortcomings result in insufficient accuracy in agent collaboration matching and low overall collaboration efficiency. Summary of the Invention

[0005] This application proposes a multi-agent collaborative interaction method based on capability matching, aiming to solve the problems of insufficient matching accuracy and low overall collaborative efficiency in existing multi-agent collaboration methods.

[0006] This application provides a multi-agent collaborative interaction method based on capability matching, the method comprising: Receive collaboration requests from the first intelligent agent; Obtain capability information of multiple candidate agents. Based on the collaboration requirements and the capability information of each candidate agent, determine the capability fit of each candidate agent relative to the collaboration requirements, the capability gap that can make up for the collaboration requirements, and the trust between the first agent and the candidate agents. The comprehensive matching score for each candidate agent is determined based on capability fit, capability gap, and credibility. Based on the comprehensive matching score, a second agent that is suitable for the collaboration needs is selected from each candidate agent, and the information of the second agent is pushed to the first agent. In response to the first intelligent agent initiating a collaboration invitation to the second intelligent agent, a collaboration relationship is established between the first and second intelligent agents, and the collaboration task corresponding to the collaboration requirement is executed based on the collaboration relationship.

[0007] Accordingly, this application also provides a multi-agent collaborative interaction device based on capability matching, comprising: The first receiving unit is used to receive the collaboration request issued by the first intelligent agent; The first acquisition unit is used to acquire the capability information of multiple candidate intelligent agents. Based on the collaboration requirements and the capability information of each candidate intelligent agent, it determines the capability adaptability of each candidate intelligent agent relative to the collaboration requirements, the capability gap that can make up for the collaboration requirements, and the trust between the first intelligent agent and the candidate intelligent agents. The first determining unit is used to determine the comprehensive matching score for each candidate agent based on the ability fit, ability gap, and credibility. The filtering unit is used to filter out the second agent that meets the collaboration requirements from each candidate agent based on the comprehensive matching score, and push the information of the second agent to the first agent. The execution unit is used to respond to the first intelligent agent initiating a collaboration invitation to the second intelligent agent, establish a collaboration relationship between the first intelligent agent and the second intelligent agent, and execute the collaboration task corresponding to the collaboration requirement based on the collaboration relationship.

[0008] Accordingly, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the program to implement the above-described multi-agent cooperative interaction method based on capability matching.

[0009] Accordingly, this application also provides a computer-readable storage medium storing a plurality of instructions adapted for loading by a processor to execute the above-described capability-matching-based multi-agent cooperative interaction method.

[0010] This application achieves fully automated interaction in a multi-agent system, from request matching to collaboration completion, by receiving a collaboration request from a first agent; acquiring capability information of multiple candidate agents; determining the capability fit of each candidate agent relative to the collaboration request, the capability gap that can be filled, and the trustworthiness between the first agent and the candidate agents, based on the collaboration request and the capability information of each candidate agent; determining a comprehensive matching score for each candidate agent based on the capability fit, capability gap, and trustworthiness; selecting a second agent suitable for the collaboration request from among the candidate agents based on the comprehensive matching score, and pushing the information of the second agent to the first agent; and establishing a collaboration relationship between the first agent and the second agent in response to the first agent initiating a collaboration invitation, and executing the collaboration task corresponding to the collaboration request based on the collaboration relationship. In this way, it realizes fully automated interaction from request matching to collaboration completion in a multi-agent system, improving the accuracy of collaboration matching and task execution efficiency. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] in: Figure 1 This is a flowchart illustrating a multi-agent collaborative interaction method based on capability matching, provided in an embodiment of this application.

[0013] Figure 2 This is a schematic diagram of the agent registration process provided in an embodiment of this application.

[0014] Figure 3 This is a schematic diagram of the capability profile preservation process provided in this application embodiment.

[0015] Figure 4 This is a schematic diagram of a capability verification process based on zero-knowledge proof, provided for an embodiment of this application.

[0016] Figure 5 This is a schematic diagram of the credit graph update and trust transfer calculation process provided in an embodiment of this application.

[0017] Figure 6 This is a schematic diagram of the complementary matching process provided in the embodiments of this application.

[0018] Figure 7 A schematic diagram of the collaborative contract state machine provided in an embodiment of this application.

[0019] Figure 8This is a schematic diagram of the overall architecture of the multi-agent collaborative interaction system provided in the embodiments of this application.

[0020] Figure 9 This is a schematic diagram of the overall process of multi-agent collaborative interaction provided in the embodiments of this application.

[0021] Figure 10 This is a structural block diagram of a multi-agent collaborative interaction device based on capability matching, provided in an embodiment of this application.

[0022] Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0023] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0024] This application provides a method, apparatus, computer-readable storage medium, and electronic device for multi-agent collaborative interaction based on capability matching. Specifically, the multi-agent collaborative interaction method based on capability matching in this application can be executed by an electronic device, which can be a terminal or a server. The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, and big data and artificial intelligence platforms.

[0025] To address the aforementioned issues, this application provides a method, apparatus, electronic device, and computer-readable storage medium for multi-agent collaborative interaction based on capability matching. This enables fully automated interaction throughout the entire process from demand matching to collaborative completion in a multi-agent system, improving the accuracy of collaborative matching and task execution efficiency.

[0026] The following sections provide detailed descriptions of each example. It should be noted that the order in which the embodiments are described is not intended to limit the preferred order of the embodiments.

[0027] This embodiment provides a multi-agent collaborative interaction method based on capability matching, which is applied to a collaborative platform server. The collaborative platform server is communicatively connected to multiple agents to receive requests and data sent by each agent, and processes, forwards, and stores them according to preset rules to establish collaborative relationships among the agents and assist in completing collaborative tasks.

[0028] Please see Figure 1 , Figure 1 This is a flowchart illustrating a multi-agent collaborative interaction method based on capability matching, provided in an embodiment of this application. Taking the application of this capability-matching-based multi-agent collaborative interaction method to a collaborative platform server as an example, the specific process can be as follows: 101. Receive the collaboration request issued by the first intelligent agent.

[0029] In this embodiment, the first intelligent agent is the initiator of the collaborative task, i.e., the requesting intelligent agent. When the first intelligent agent needs the assistance of other intelligent agents to complete a certain task, it can send a collaboration request to the collaboration platform server. The collaboration request may include request description information (such as request description text), filtering constraints, etc.

[0030] After receiving the collaboration request, the collaboration platform server temporarily stores the request and triggers the subsequent matching process.

[0031] In some embodiments, prior to the step of "receiving the collaboration request published by the first agent", the method may further include an agent registration process, as follows: In response to the registration request of the intelligent agent, the system receives the public key submitted by the intelligent agent, generates a decentralized identity identifier for the intelligent agent based on the public key, and stores the correspondence between the decentralized identity identifier and the public key to complete the registration of the intelligent agent. The intelligent agent includes at least a first intelligent agent and a second intelligent agent.

[0032] Specifically, when each agent first connects to the collaboration platform, it generates a public-private key pair locally (e.g., using the ED25519 or Secp256k1 algorithm). The private key is kept safe by the agent itself; the platform does not access or store the private key. The agent then initiates a registration request to the collaboration platform server based on the generated public key.

[0033] In response to a registration request from an agent, the collaboration platform server receives the agent's public key and optional metadata. Upon receiving the public key, the server first verifies its format to ensure it conforms to the relevant cryptographic algorithm specifications. After successful verification, the server calculates and generates the agent's decentralized identity (DID) based on the public key, in the format did:agent: <hash>,in <hash>This is the hash value calculated based on the public key content.

[0034] After generating the DID, the server checks if the DID already exists in the database. If it does not exist, the server associates the DID with the corresponding public key, registration time, and other metadata, completing the agent's registration. The server then returns the generated DID to the agent.

[0035] Through the above process, the first and second intelligent agents completed their registration on the collaboration platform and obtained their own independent public-private key pairs and decentralized identity identifiers.

[0036] After an agent completes registration, any request sent to the collaboration platform must include its DID and a digital signature generated based on its private key, so that the server can verify the legitimacy and integrity of the request. These requests include, but are not limited to: file updates, collaboration request publication, collaboration invitations, and contract signing.

[0037] For example, please see Figure 2 , Figure 2 This is a schematic diagram of the agent registration process provided in an embodiment of this application. Figure 2 As shown, when an intelligent agent first accesses the collaboration platform, it needs to complete identity registration in order to obtain a decentralized identity identifier in the collaboration platform.

[0038] After initiating registration, the agent first generates an asymmetric key pair locally, including a public key and a private key. The private key is kept secret by the agent and is not sent to the collaboration platform server. The agent submits the generated public key to the collaboration platform server as part of the registration request.

[0039] After receiving the public key submitted by the agent, the collaboration platform server first verifies the public key format to determine if it conforms to the corresponding cryptographic algorithm specifications. If the public key format is incorrect, the server returns an error message to the agent, prompting it to resubmit the correct public key. If the public key format is correct, the server calculates a hash value based on the public key and generates a decentralized identity (DID) for the agent based on this hash value, for example, using did:agent: <hash>The format.

[0040] After generating the DID, the server checks if the DID already exists in the database. If it does, it means the agent has completed registration, and the server can return a registration confirmation message to the agent. If it does not exist, the server associates the DID with the corresponding public key, registration timestamp, and other metadata and stores it in the database, thus completing the agent's registration.

[0041] After registration is complete, the server returns the generated DID to the agent. The agent saves the DID and includes it in all subsequent requests that interact with the platform, so that the platform can use the DID to query the corresponding public key for signature verification.

[0042] In some embodiments, the step "receiving a collaboration request published by the first agent" may include the following operations: Receive collaboration requests, which carry the decentralized identity of the first intelligent agent and a first digital signature generated based on the first intelligent agent's private key; The public key is retrieved based on the decentralized identity of the first agent, and the validity of the first digital signature is verified using the public key. If the verification passes, the collaboration request will be published.

[0043] In this embodiment, the first intelligent agent uses the corresponding private key to digitally sign the content of the collaboration request, generates a first digital signature, and sends the collaboration request carrying the decentralized identity identifier, the content of the collaboration request, and the first digital signature to the collaboration platform server.

[0044] After receiving a collaboration request, the server queries the corresponding public key based on the decentralized identity identifier and uses that public key to verify the validity of the first digital signature. If the verification passes, it means that the request indeed originated from the first intelligent agent and its content has not been tampered with, and the server publishes the collaboration request; if the verification fails, the request is rejected.

[0045] Through the above verification mechanism, the collaboration platform server can authenticate and verify the integrity of all subsequent requests, ensuring the security and non-repudiation of the interaction.

[0046] In some embodiments, prior to the step of "receiving the collaboration request issued by the first intelligent agent", the method may further include a capability profile submission and authentication process, as follows: Receive capability profiles submitted by intelligent agents. The capability profiles carry the decentralized identity of the intelligent agents and a digital signature generated based on the intelligent agents' private keys. The public key is retrieved based on the decentralized identity identifier, and the validity of the digital signature is verified using the public key. If the verification passes, the competency profile will be validated. After verification, the capability profile is stored in a decentralized storage network, the content address corresponding to the capability profile is obtained, and the correspondence between the decentralized identity and the content address is stored. The intelligent agent includes a first intelligent agent and a second intelligent agent.

[0047] After registration, an agent can submit its capability profile to the collaboration platform server. The capability profile is structured data describing the agent's various capabilities, and can be in JSON format. It includes information such as execution capability descriptions, skill tags, quantified experience data, and project history.

[0048] Specifically, when submitting a capability profile, the agent uses its private key to digitally sign the content of the capability profile and sends a submission request to the collaboration platform server, carrying the agent's decentralized identity, the content of the capability profile, and the digital signature.

[0049] After receiving the capability profile, the collaboration platform server first queries the corresponding public key based on the agent's decentralized identity and uses this public key to verify the validity of the digital signature. If the verification passes, it means that the capability profile does indeed originate from the claimed agent and its content is complete and has not been tampered with; if the verification fails, the server rejects the submission.

[0050] After successful signature verification, the server verifies the capability profile, including checking whether it conforms to the preset JSON Schema definition and whether required fields are complete. If the verification fails, the server returns a verification failure message to the agent.

[0051] After successful verification, the server calculates the content hash value of the capability profile, uploads the capability profile to the decentralized storage network, obtains the returned content address, and stores the mapping between the agent's decentralized identity and the content address. Since the content address is generated based on the profile content, any modification to the profile content will result in a change to the content address, thus ensuring the immutability of the profile content.

[0052] Through the above process, both the first and second intelligent agents can submit their capability profiles to the collaboration platform server, and after signature verification and integrity verification, the profiles are stored in a decentralized storage network.

[0053] In some embodiments, the step "storing the capability profile to a decentralized storage network and obtaining the content address corresponding to the capability profile" may include the following operations: The hash value of the computing capability profile is used to upload the capability profile to a decentralized storage network, and the content address returned by the decentralized storage network is received.

[0054] The collaboration platform server uses a preset hash algorithm (such as SHA-256) to calculate the hash value of the verified capability file content. This hash value uniquely identifies the file content, and any modification to the file content will cause the hash value to change.

[0055] The collaboration platform server uploads capability files to a decentralized storage network (such as IPFS). After receiving and storing the capability file, the decentralized storage network generates a corresponding Content Identifier (CID) based on the file content and returns this CID to the collaboration platform server. Capability files with the same content will receive the same CID, while files with different content will receive different CIDs.

[0056] Finally, the collaboration platform server stores the correspondence between the agent's decentralized identity identifier and the CID in the local database, so that the corresponding capability profile content address can be queried later based on the DID.

[0057] Through the above method, the capability profile is stored in a decentralized storage network, and its content address is generated based on the profile content itself, which makes the profile content verifiable, complete and tamper-proof.

[0058] In some embodiments, the server further constructs a retrieval index based on the content of the capability profile and selectively submits the profile's evidence information to the blockchain. The method may also include the following steps: Extract skill tags from the ability profile and update the inverted index based on the skill tags; Feature extraction is performed on the competency profile to obtain competency feature vectors, which are then stored in the competency feature vector library. Submit the content address and / or content hash value to the blockchain for evidence storage.

[0059] Specifically, the server extracts skill tags from the capability profile and updates the inverted index based on the extracted skill tags to support subsequent fast retrieval based on skill tags. At the same time, the server performs feature extraction on the capability profile (such as calling the BERT encoder to convert the profile text into a 768-dimensional floating-point vector) to obtain capability feature vectors, which are stored in a capability feature vector library (such as the Milvus vector database) for subsequent matching engine to perform approximate nearest neighbor retrieval.

[0060] In this application, the matching engine is a core functional module in the collaboration platform server responsible for agent discovery and recommendation. It is used to select the most suitable and trustworthy second agent from multiple registered candidate agents based on the collaboration needs published by the first agent. Furthermore, the server can also submit the content hash value and / or content address of the capability profile to the blockchain for notarization, for example, by batch packaging multiple notarized information entries and submitting them to the blockchain network via a smart contract.

[0061] For example, please see Figure 3 , Figure 3 This is a schematic diagram illustrating the capability profile storage process provided in this application embodiment. For example... Figure 3 As shown, when submitting a capability profile, the agent uses its private key to digitally sign the content of the capability profile and sends a submission request carrying the agent's decentralized identity, the content of the capability profile, and the digital signature to the collaboration platform server.

[0062] After receiving the capability profile submitted by the agent, the collaboration platform server first queries the corresponding public key based on the agent's decentralized identity and uses this public key to verify the validity of the digital signature. If the signature verification fails, the server rejects the submission. If the signature verification passes, the server performs schema validation on the capability profile, checking whether it conforms to the preset JSON schema definition and whether required fields are complete. If the schema validation fails, the server rejects the submission. If the validation passes, the server calculates the content hash value (e.g., SHA-256) of the capability profile, uploads the capability profile to a decentralized storage network (e.g., IPFS), and obtains the corresponding content address (CID) of the profile. The server then stores the mapping between the agent's decentralized identity and the content address in a PostgreSQL database.

[0063] In some implementations, the server can also submit the content hash value and / or content address of the capability profile to the blockchain for notarization (represented as "optional on-chain" in the diagram), serving as a trusted credential that the profile existed at a specific point in time and that the content has not been tampered with.

[0064] Simultaneously, the server extracts features from the capability profile (e.g., by calling the BERT encoder to convert the profile text into a 768-dimensional floating-point vector), obtaining the capability feature vector of the agent. This capability feature vector is stored in the Milvus vector database for subsequent matching engine to perform approximate nearest neighbor retrieval. The server also extracts skill tags from the capability profile and updates the Elasticsearch inverted index based on the extracted skill tags to support subsequent fast retrieval based on skill tags.

[0065] In some embodiments, the above-described step of verifying the capability profile may further include a capability authenticity verification process based on zero-knowledge proofs. This process is used to verify whether the capabilities claimed by the agent are true and credible without obtaining the agent's underlying model; that is, the step "verifying the capability profile" may include the following operations: Obtain the test dataset and verification standards associated with the capability type corresponding to the capability profile; Receive zero-knowledge proofs submitted by the agent. The zero-knowledge proof is generated locally by the agent based on the test dataset and the agent's model. It is used to prove that the model's test results on the test dataset meet the verification criteria. Call the validator to verify the zero-knowledge proof; If the verification passes, a verified mark is added to the capability profile, and the proof address of the zero-knowledge proof is stored.

[0066] Specifically, the collaboration platform server pre-associates corresponding test datasets and validation standards for different capability types and publishes them to each agent. The test datasets are publicly available standard datasets used to uniformly evaluate the model performance of each agent; the validation standards include performance thresholds (e.g., image classification accuracy ≥90%, inference speed ≤100ms, etc.) and description files of the evaluation circuits (e.g., circuits written in the Circom language). The server can also publish the hash values ​​of the test datasets, allowing agents to verify the integrity of the datasets after downloading them, ensuring that the test data has not been tampered with.

[0067] After submitting its capability profile, if an agent wishes to verify the authenticity of a specific capability, it first retrieves the test dataset and verification criteria associated with that capability type from the server. The agent loads its model locally, inputs the test dataset into the model for inference, and obtains the model's output. Subsequently, based on the model weights (as private input), the test dataset, and the model output (as public input), the agent runs a zero-knowledge proof generation program (such as using the SNARK.js library) to generate a local zero-knowledge proof. This zero-knowledge proof demonstrates to the verifier that the model's performance on the test dataset meets the preset verification criteria without revealing the model weights, model structure, or any internal parameters.

[0068] After generating a zero-knowledge proof, the agent submits it to the collaboration platform server. The server then invokes a validator to verify the proof. The validator can be a smart contract deployed on the blockchain (on-chain validator) or a cryptographic library deployed locally on the server (off-chain validator). The validator performs cryptographic operations on the proof according to pre-published verification standards and evaluation circuits to determine its validity.

[0069] If the verification passes, it means that the agent's model performance on the test dataset does indeed meet the preset verification criteria, indicating that the agent's claimed capabilities are genuine and credible. The server adds a "verified" tag (e.g., capability_verified: true) to the agent's capability profile and stores the proof address of the zero-knowledge proof (e.g., IPFS storage address or blockchain transaction hash), recording information such as the verification time for subsequent verification. Afterward, other agents viewing the capability profile will see the "verified" badge, thereby increasing their trust in the agent's capabilities.

[0070] In some implementations, the server can also periodically trigger the re-verification of verified capabilities (such as requiring the agent to regenerate and submit proofs at regular intervals) to prevent verified capabilities from no longer meeting the verification criteria due to model updates or performance degradation, thereby ensuring the continued validity of the "verified" mark in the capability profile.

[0071] For example, please see Figure 4 , Figure 4 This is a schematic diagram of a capability verification process based on zero-knowledge proof, provided for an embodiment of this application. Figure 4 As shown, the collaboration platform server pre-discloses the test dataset and evaluation circuit associated with this capability type. The test dataset is a publicly available standard dataset used to uniformly evaluate the model performance of each agent; the evaluation circuit (e.g., written in Circom language) defines the verification logic and verification criteria (e.g., image classification accuracy ≥ 90%).

[0072] After acquiring the test dataset, the agent loads its private model locally, inputs the test dataset into the model for inference, and obtains the model's output. Subsequently, the agent uses the model weights as private input and the test dataset and model output as public input to run a zero-knowledge proof generation program, generating a zero-knowledge proof locally. This proof demonstrates to the verifier that the model's performance on the test dataset meets preset verification criteria without revealing the model weights and structure.

[0073] After generating the proof, the agent submits it to the collaboration platform server. The server then invokes a validator to verify the proof. In some implementations, the validator can be deployed as a smart contract on the blockchain (on-chain contract), leveraging the blockchain's immutability and transparency to make the verification process and results publicly auditable. The server calls the on-chain contract's verification interface, passing in the proof data, and the on-chain contract executes the verification logic and returns the verification result.

[0074] If the verification passes, it means that the agent's model performance on the test dataset does indeed meet the preset verification criteria, indicating that the agent's claimed capabilities are credible. The server adds a "verified" tag to the agent's capability profile and stores the verification address for subsequent verification. If the verification fails, the server marks it as rejected, and the capability profile does not receive the "verified" tag.

[0075] 102. Obtain the capability information of multiple candidate intelligent agents. Based on the collaboration requirements and the capability information of each candidate intelligent agent, determine the capability adaptability of each candidate intelligent agent relative to the collaboration requirements, the capability gap that can make up for the collaboration requirements, and the credibility between the first intelligent agent and the candidate intelligent agents.

[0076] In this embodiment, the collaboration platform server obtains capability information from multiple registered candidate agents. Capability information may include execution capability descriptions, skill tags, quantified experience data, and project history recorded in the capability profile.

[0077] The collaboration platform server can evaluate each candidate agent based on the collaboration requirements and their capabilities, using the following three dimensions: Capability Fit: This dimension measures the degree of match between the candidate agent's overall capabilities and the collaboration requirements. Capability Gap: This dimension measures the extent to which the candidate agent can compensate for the capabilities required in the collaboration. Trustworthiness: This dimension measures the level of trust between the first agent and the candidate agents.

[0078] In some embodiments, the step "based on the collaboration requirements and the capability information of each candidate agent, determining the capability fit of each candidate agent relative to the collaboration requirements, the capability gap that can compensate for the collaboration requirements, and the trustworthiness between the first agent and the candidate agents" may include the following operations: Feature extraction is performed on the collaboration requirements to obtain the requirement feature vector, and feature extraction is performed on the capability information of each candidate agent to obtain the capability feature vector. The similarity between the requirement feature vector and the capability feature vector is calculated as the capability fit. Extract the set of skill requirements from the collaboration needs, obtain the set of skill tags for each candidate agent, match the set of skill requirements with the set of skill tags, and determine the capability gap based on the matching results; Obtain a pre-constructed reputation graph, calculate the trust transfer score between the first agent and the candidate agents based on the reputation graph, and determine the credibility based on the trust transfer score.

[0079] In some embodiments, the collaboration platform server first extracts features from the collaboration requirements and converts them into a numerical vector representation, namely, a requirement feature vector. Specifically, for requirement text described in natural language, the server can use a pre-trained language model such as Sentence-BERT to encode the requirement text into a 768-dimensional floating-point vector. Simultaneously, for the capability information of each candidate agent, the server also uses the same encoding method (such as Sentence-BERT) to convert it into capability feature vectors.

[0080] After obtaining the requirement feature vector and the capability feature vectors of each candidate agent, the server calculates the similarity (e.g., cosine similarity) between the requirement feature vector and each capability feature vector, and uses the calculated similarity as the capability fit. The higher the similarity, the better the overall capability of the candidate agent matches the collaboration requirements.

[0081] In some embodiments, the collaboration platform server first extracts a set of skill requirements from the collaboration needs, which includes one or more skill tags required to complete the needs. Simultaneously, the server obtains a set of skill tags for each candidate agent, which is recorded in the candidate agent's capability profile.

[0082] The server matches the set of skill requirements with the set of skill tags for each candidate agent, and determines the capability gap for each candidate agent based on the matching results. Specifically, if a candidate agent possesses a skill from the set of skill requirements, the corresponding gap is filled; if the candidate agent does not possess a skill, the gap remains unfilled. The server calculates the coverage of the skill requirement set by each candidate agent, and determines the capability gap accordingly.

[0083] The relationship between the matching result and the capability gap can be determined in several ways. For example, the server can determine the capability gap as the ratio of the intersection of the skill requirement set and the skill tag set to the size of the skill requirement set. The larger the ratio, the more capability gaps the candidate agent can fill. Alternatively, the server can determine the capability gap as the size of the difference between the skill requirement set and the skill tag set. The smaller the difference, the better the candidate agent matches the skill requirements of the collaboration needs.

[0084] In some embodiments, a reputation graph is pre-built in the collaboration platform server. This reputation graph uses decentralized identities of multiple agents as nodes and historical mutual ratings between agents as edge weights, forming a directed weighted graph reflecting the trust relationships between agents. The server calculates the trust transfer score between the first agent and each candidate agent based on this reputation graph, and uses the calculated trust transfer score as the credibility.

[0085] In some embodiments, the method may further include the following steps: By using the decentralized identity identifiers of multiple agents as nodes and the historical mutual rating scores between agents as edge weights, a directed weighted graph is constructed to obtain the reputation graph.

[0086] In some embodiments, the collaboration platform server maintains a reputation graph to record and calculate trust relationships between agents. The reputation graph is a directed weighted graph, where nodes are decentralized identity identifiers (DIDs) of each agent, edges represent an agent's trust rating of another agent, the direction of the edge is from the evaluator to the evaluated agent, and the weight of the edge is the historical mutual rating score.

[0087] Initially, newly registered agents have not yet participated in any collaborations, so there are no edges related to that agent in the reputation graph. As agents complete collaborative tasks, they rate each other based on their collaborative experience, and the reputation graph gradually becomes richer.

[0088] In some embodiments, the method may further include the following steps: Receive the mutual score in the collaboration completion event, and update the edge weights in the reputation graph based on the mutual score.

[0089] In some embodiments, whenever a collaborative task between the first and second agents is completed, both agents call a scoring interface to submit a mutual score to the collaboration platform server. The mutual score can be a value from 0 to 5, with a higher score indicating greater satisfaction with the other agent.

[0090] After receiving the mutual rating scores from the collaboration completion event, the collaboration platform server records the rating party, the rated party, the score, and the rating timestamp in the trust score table. Simultaneously, the server triggers an incremental update to the reputation graph: it searches the in-memory graph for a directed edge from the rating party to the rated party. If the edge already exists, its weight is updated based on the new mutual rating score (e.g., calculating a weighted average or moving average of historical mutual rating scores). If the edge does not exist, a new directed edge is added to the reputation graph from the rating party to the rated party, using the current mutual rating score as the edge weight.

[0091] In some embodiments, the step "calculating the trust transfer score between the first agent and the candidate agents based on the reputation graph" may include the following operations: Using the first agent as the root node, perform random walk iterative calculations in the reputation graph until the convergence condition is met. Based on the iterative calculation results, determine the trust transfer score between the first agent and the candidate agents.

[0092] In some embodiments, the server employs the Personalized PageRank algorithm to calculate the trust transfer score. The core idea of ​​this algorithm is to simulate a random walk process starting from the root node in the reputation graph, with the first agent as the root node. Each step involves the walker jumping along the directed edge of the current node to an adjacent node with a certain probability, or jumping back to the root node with a certain probability. After multiple iterations, the probability of each node being visited gradually converges to a stable value, which represents the trust transfer score from the root node to each node.

[0093] Specifically, when the matching engine requests to calculate the trust transfer score from the first agent to the candidate agents, the collaboration platform server first checks if a recently calculated Personalized PageRank result exists in the cache. If the cache exists and has not expired, the trust transfer score in the cache is returned directly. If the cache does not exist or has expired, the server starts iterative calculation of Personalized PageRank in the reputation graph, with the first agent as the root node. Based on the access probability of each node after the iterative calculation converges, the server determines the trust transfer score from the first agent to each candidate agent and returns it to the matching engine.

[0094] In some implementations, the server can also periodically (e.g., hourly) pre-calculate Personalized PageRank scores for all active agent nodes and cache the results to speed up the response time when the matching engine is called later.

[0095] For example, please see Figure 5 , Figure 5 A schematic diagram illustrating the reputation graph update and trust transfer calculation process provided in this application embodiment. See also... Figure 5 After the collaborative task is completed, the first and second agents submit their mutual evaluation scores (e.g., scores for each other) to the collaborative platform server. Figure 5 (As shown in the example "Agent A scores B, Agent B scores A"). After receiving the mutual scores submitted by both parties, the server associates and stores the scorer, the scored party, the score, and the scoring timestamp in the trust score table.

[0096] After storage is complete, the server triggers an incremental update of the reputation graph, adding or updating the corresponding directed edge weights (e.g., ...) in the memory graph. Figure 5 (As shown in the "Update Memory Graph"). When the matching engine needs to obtain the trust transfer score from the first agent to the candidate agents, it sends a request to the reputation graph service (e.g., ...). Figure 5 (As shown in "Matching Engine Request"). The reputation graph service first queries the cached trust matrix. If the cache is hit, the cached trust transfer score is directly obtained. If the cache is not hit, the server starts Personalized PageRank incremental calculation with the first agent as the root node, performing random walk iterations on the reputation graph until convergence. Based on the convergence result, the trust transfer score from the first agent to each candidate agent is determined. After the calculation is completed, the service stores the result in the cache and returns it to the matching engine (e.g., ...). Figure 5 (As shown in "Return trust(A, B)").

[0097] 103. Determine the comprehensive matching score for each candidate agent based on capability fit, capability gap, and credibility.

[0098] In this embodiment, the collaboration platform server fuses the evaluation results of the three dimensions determined in the above steps to obtain a comprehensive matching score for each candidate agent. This comprehensive matching score is used to reflect the overall suitability of each candidate agent as a collaboration partner of the first agent.

[0099] In some embodiments, the step "determine the comprehensive matching score for each candidate agent based on capability fit, capability gap, and credibility" may include the following operations: The ability fit, ability gap and credibility are weighted and summed, and the comprehensive matching score is determined based on the weighted summation result.

[0100] The weights corresponding to each dimension can be preset or dynamically adjusted according to the actual application scenario. For example, in scenarios that require strong capability matching, the weight of capability adaptability can be increased, and in scenarios that require high-trust collaboration, the weight of credibility can be increased.

[0101] The calculation formula can be expressed as: Complement_Score =α·Fitness +β·Gap +γ·Trust; In this context, Fitness represents capability fit, Gap represents capability gap, and Trust represents credibility. α, β, and γ are the preset weight coefficients for their respective dimensions. The magnitude of these weight coefficients reflects the relative importance of each dimension in the overall matching process. For example, in scenarios where it's crucial to ensure the candidate agent can fulfill the task requirements, the weight α for capability fit can be set higher; conversely, in scenarios where it's essential to ensure sufficient trust between the requester and the candidate agent, the weight γ for credibility can be set higher.

[0102] By using the weighted summation method described above, the matching engine can quantify the evaluation indicators of three different dimensions into a comprehensive matching score. This score comprehensively reflects the overall suitability of each candidate agent relative to the collaborative needs, providing a quantitative basis for subsequent agent selection and recommendation.

[0103] 104. Based on the comprehensive matching score, select the second agent that is suitable for the collaboration needs from each candidate agent, and push the information of the second agent to the first agent.

[0104] In this embodiment, the collaboration platform server sorts or filters the candidate agents based on their comprehensive matching scores, and determines one or more candidate agents that are suitable for the collaboration requirements as recommended results. For ease of description, this embodiment refers to the finally selected suitable candidate agents as the second agent.

[0105] The collaboration platform server pushes the information of the second agent (such as the second agent's DID, capability profile summary, comprehensive matching score, etc.) to the first agent for the first agent to view and select.

[0106] In some embodiments, the method may further include the following steps: In the candidate agent capability feature vector library, retrieve multiple candidate agents whose matching degree with the requirement feature vector meets the preset conditions; Based on the comprehensive matching score, a second agent suitable for the collaborative requirements is selected from each candidate agent, including: Obtain the screening constraints in the collaboration requirements, and filter each candidate agent according to the screening constraints; From the filtered candidate agents, the candidate agents whose comprehensive matching scores meet the preset requirements are selected as the second agent.

[0107] In some embodiments, to improve matching efficiency, the collaboration platform server can first perform an initial screening to quickly select a candidate subset related to collaboration needs from a large number of registered agents.

[0108] Specifically, the server extracts features from the collaboration requirements to obtain requirement feature vectors. Then, it retrieves the top N agents (e.g., N=500) with the highest matching degree between the capability feature vectors and the requirement feature vectors from the candidate agent capability feature vector library, which serve as multiple candidate agents after initial screening. In some implementations, the capability feature vector library can use the Milvus vector database, and fast retrieval can be achieved through approximate nearest neighbor search algorithms (such as IVF-PQ, HNSW, etc.).

[0109] After obtaining the comprehensive matching score of each candidate agent, the server retrieves the screening constraints in the collaboration requirements (such as minimum success rate, maximum response time, etc.), and filters each candidate agent according to these constraints, excluding those that do not meet any of the constraints. Then, from the remaining candidate agents after filtering, the server selects the candidate agents whose comprehensive matching scores meet the preset requirements as the second agent, for example, selecting the one with the highest comprehensive matching score, or selecting multiple candidate agents ranked in the top K (e.g., K=20) for further selection.

[0110] For example, please see Figure 6 , Figure 6 This is a schematic diagram of the complementary matching process provided in an embodiment of this application. Figure 6 As shown, the matching engine first obtains the requirement text and hard constraints from the collaboration request. Hard constraints can include filtering thresholds such as minimum success rate and maximum response time.

[0111] Subsequently, the matching engine calls the BERT encoder to encode the request text, converting it into a fixed-dimensional request vector. In some implementations, the BERT encoder can use the Sentence-BERT model to convert the request text into a 768-dimensional floating-point vector.

[0112] After obtaining the demand vector, the matching engine performs an approximate nearest neighbor search in the Milvus vector database, retrieving the 500 candidate agents with the largest 1-cosine_similarity (cosine distance) to the demand vector, which are then used as the candidate set after initial screening.

[0113] After obtaining the initial candidate set, the matching engine iterates through each candidate agent in the set and performs the following calculations for each candidate: It retrieves the candidate agent's skill tag set from Elasticsearch and compares it with the skill requirement set in the collaboration needs to calculate a skill gap score; it also retrieves the trust transfer value from the first agent to the candidate agent from the reputation graph service. The matching engine then calculates the candidate's complementarity score according to a preset formula based on semantic differences, skill gaps, and trust transfer values.

[0114] After calculating the complementarity score, the matching engine filters and judges the candidate based on hard constraints (such as...). Figure 6 (See "Meets hard constraints?"). If a candidate does not meet any hard constraint (e.g., success rate is below a threshold), the candidate is discarded. If a candidate meets all hard constraints, the candidate and its complementarity score are retained.

[0115] Finally, the matching engine sorts all the retained candidates in descending order of complementarity score and returns the top-K candidate agents as recommendations for the first agent to choose from.

[0116] 105. In response to the first intelligent agent initiating a collaboration invitation to the second intelligent agent, a collaboration relationship is established between the first intelligent agent and the second intelligent agent, and the collaboration task corresponding to the collaboration requirement is executed based on the collaboration relationship.

[0117] In this embodiment, after the first intelligent agent views the pushed information of the second intelligent agent, if it confirms that it has selected the second intelligent agent as the collaboration object, it sends a collaboration invitation to the collaboration platform server, which is directed to the second intelligent agent.

[0118] In response to the collaboration invitation, the collaboration platform server establishes a collaboration relationship between the first and second intelligent agents, including but not limited to: establishing a negotiation channel for both parties to negotiate collaboration terms, generating a collaboration contract based on the negotiation results, receiving and storing the signed confirmations of the collaboration contract from both parties, providing message channels and log records during task execution, and triggering payment based on acceptance confirmation, etc.

[0119] Once the collaborative relationship is established, the first and second agents execute the collaborative tasks corresponding to the collaborative requirements based on the collaborative relationship until the tasks are completed.

[0120] In some embodiments, the step "in response to the first agent initiating a collaboration invitation to the second agent, establishing a collaboration relationship between the first agent and the second agent, and executing the collaboration task corresponding to the collaboration requirement based on the collaboration relationship" may include the following operations: In response to the first intelligent agent sending a cooperation invitation to the second intelligent agent, a negotiation channel is established between the first intelligent agent and the second intelligent agent to allow the first intelligent agent and / or the second intelligent agent to negotiate the cooperation terms. In response to the first and second intelligent agents reaching an agreement on the negotiated cooperation terms, a cooperation contract is generated based on the negotiated cooperation terms. Receive the first signature of the cooperation contract by the first intelligent agent based on the first private key and the second signature of the cooperation contract by the second intelligent agent based on the second private key, and store the cooperation contract, the first signature and the second signature; During the execution of collaborative tasks, encrypted message channels are provided for the first and second agents, and execution logs are recorded. In response to the acceptance confirmation instruction from the first intelligent agent, a payment operation is triggered according to the payment conditions in the collaboration contract, transferring resources from the first intelligent agent's account to the second intelligent agent's account.

[0121] The first agent sends a collaboration invitation to the second agent. This invitation may include collaboration terms such as task description, token budget, and deadline. The collaboration platform server responds to the invitation by establishing a negotiation channel between the first and second agents. This channel allows for multiple rounds of communication and modification of the collaboration terms—the second agent can accept, reject, or propose modifications (such as adjusting the budget or milestones). The two agents interact through the negotiation channel until they reach an agreement on the collaboration terms.

[0122] After both parties reach an agreement, the collaboration platform server generates a collaboration contract based on the negotiated collaboration terms. The contract includes information such as the DIDs of both parties, task milestones, deliverable standards, and payment terms. The first agent signs the contract using its first private key, and the second agent signs the contract using its second private key. Both parties submit their signatures to the platform. After verifying the validity of the signatures, the platform associates and stores the contract and the signatures of both parties, thus making the contract effective.

[0123] Once the contract takes effect, the task execution phase begins. The collaboration platform server provides an encrypted messaging channel for both parties to ensure communication security, and also records execution logs (including timestamps, message hashes, and other metadata) for later traceability.

[0124] Upon completion of the task, the first agent confirms acceptance and sends an acceptance confirmation instruction to the platform. In response, the platform triggers a payment operation based on the payment conditions in the contract (e.g., calling an external payment system via a payment adapter), transferring the agreed amount of tokens from the first agent's account to the second agent's account, and recording the collaboration completion event for the reputation graph service to update mutual score ratings.

[0125] For example, please see Figure 7 , Figure 7 A schematic diagram of the collaborative contract state machine provided in an embodiment of this application.

[0126] like Figure 7 As shown, after the first agent selects the second agent as the collaboration partner, it calls the creation invitation interface to send a collaboration invitation to the collaboration platform server. At this time, the contract enters the initial state: draft state.

[0127] The collaboration platform server generates an invitation ID, sends the collaboration invitation to the second intelligent agent, and updates the contract status to pending, waiting for the second intelligent agent's response.

[0128] Upon receiving a collaboration invitation, the second agent can accept, modify, or reject it. If the second agent rejects the invitation or fails to respond within a preset time, the contract enters a timeout / rejection state, and the negotiation process terminates. If the second agent accepts the invitation, or proposes modifications to the collaboration terms and reaches an agreement with the first agent, the contract enters a negotiated state, indicating that both parties have reached an agreement on the collaboration terms.

[0129] In the negotiated state, the collaboration platform server generates a collaboration contract based on the negotiated terms and sends the contract to both the first and second agents for signing. Both parties sign the contract using their respective private keys and submit it to the platform. If one party fails to sign or refuses to sign, the contract enters the negotiation failure state, and the process terminates. If both parties complete the signing and the platform verifies it, the contract enters the signed state.

[0130] After the contract is signed, the first and second agents begin executing the collaborative task, and the contract status is updated to "executing". During task execution, the two parties exchange data and intermediate results through an encrypted message channel provided by the collaboration platform server, and the platform records execution logs.

[0131] After the task is completed, the first agent verifies and accepts the delivered results. The platform then triggers a payment operation based on the verification instruction, transferring tokens from the first agent's account to the second agent's account. Once verification and payment are complete, the contract enters the "completed" state, indicating that the collaboration is fully finished.

[0132] This application discloses a multi-agent collaborative interaction method based on capability matching. The method includes: receiving a collaboration request published by a first agent; acquiring capability information of multiple candidate agents; determining, based on the collaboration request and the capability information of each candidate agent, the capability fit of each candidate agent relative to the collaboration request, the capability gap that can compensate for the collaboration request, and the trustworthiness between the first agent and the candidate agents; determining a comprehensive matching score for each candidate agent based on the capability fit, capability gap, and trustworthiness; selecting a second agent from among the candidate agents that matches the collaboration request based on the comprehensive matching score, and pushing the information of the second agent to the first agent; in response to the first agent initiating a collaboration invitation to the second agent, establishing a collaboration relationship between the first agent and the second agent, and executing the collaboration task corresponding to the collaboration request based on the collaboration relationship. This achieves fully automated interaction from request matching to collaboration completion in a multi-agent system, improving the accuracy of collaboration matching and task execution efficiency.

[0133] Based on the above description, the following examples will further illustrate the capability-matching-based multi-agent collaborative interaction method of this application. Please refer to... Figure 8 , Figure 8 This is a schematic diagram of the overall architecture of the multi-agent collaborative interaction system provided in this application embodiment. The multi-agent collaborative interaction system provided in this application embodiment includes a platform function layer and an agent layer. In the platform function layer, the collaborative platform server provides the following core service modules: The registration and identity service generates decentralized identity identifiers (DIDs) for each agent accessing the platform, manages the public keys of each agent, and verifies the signatures of all subsequent requests. All requests issued by each agent (such as file submissions, requirement publications, contract signings, etc.) must carry a digital signature generated based on its private key. This service queries the corresponding public key based on the DID in the request for verification, ensuring the authenticity of the request source and the integrity of the content.

[0134] The Capability Profile Service manages the four-dimensional capability profiles of each agent, including profile creation, updating, storage, notarization, retrieval, and display. Profile content is stored in a decentralized storage network after signature verification and format validation, and the mapping between DID and content address is notarized in a database for querying by other agents or matching engines.

[0135] The zero-knowledge proof capability verification service is used to verify the authenticity of the capabilities claimed by an agent. The service provides standardized test datasets and evaluation circuits for specific capability types. The agent runs the test set locally and generates zero-knowledge proofs. The platform calls the verifier to verify the proofs, and adds a "verified" mark to the capability profile after successful verification.

[0136] The collaboration contract service manages the entire process of collaboration between intelligent agents, including sending and receiving collaboration invitations, negotiation between the parties, contract generation and signing, task execution support, acceptance confirmation, and resource settlement. After both parties reach an agreement on the collaboration terms, the service generates a collaboration contract and it becomes effective after being signed by both parties' private keys; after the task is completed, the service triggers the payment operation according to the acceptance confirmation instruction.

[0137] At the agent layer, each agent completes registration, identity authentication, capability verification, collaboration matching, and task execution through the aforementioned services. Collaboration records between agents are stored and used for subsequent reputation updates.

[0138] For further details, please refer to Figure 9 , Figure 9 This is a schematic diagram of the overall process of multi-agent collaborative interaction provided in the embodiments of this application.

[0139] The multi-agent collaborative interaction method based on capability matching provided in this application involves the interaction between a first agent (demand party), a collaborative platform server, a second agent (partner), and an external payment system. The following is combined with... Figure 9 The entire process will be explained in detail.

[0140] When agent A (the requester) and agent B (the collaborator) first connect to the collaboration platform, they each send a registration request to the platform server, submitting their respective public keys. The platform server generates a corresponding decentralized identity (DID) based on each agent's public key, stores the mapping between the DID and the public key, and returns the DID to each agent, completing the registration.

[0141] After registration, Agent A and Agent B submit their respective capability profiles to the collaboration platform server. After signature verification and format validation, the capability profiles are stored in a decentralized storage network (such as IPFS) to obtain a Content Address (CID), and the correspondence between the CID and DID is stored in a database. Optionally, the content hash value and / or CID of the profile can also be submitted to the blockchain for notarization.

[0142] Agent A (the requester) publishes a collaboration request to the collaboration platform server. This request includes the request text, budget, and screening constraints (such as minimum success rate, maximum response time, etc.). Upon receiving the collaboration request, the collaboration platform server invokes the matching engine to execute a complementary matching process. The matching engine first retrieves the top-performing candidate agents from the capability feature vector library based on the request text as initial screening results. For each candidate agent in the initial screening results, the matching engine calculates capability fit, capability gap, and credibility, and integrates these three dimensions into a comprehensive matching score. Subsequently, the matching engine filters the candidate agents according to the screening constraints, sorts them in descending order of comprehensive matching score, and returns the Top-K candidate list to Agent A.

[0143] Agent A selects Agent B from the candidate list as a collaboration partner and sends a collaboration invitation, which includes task details, to the collaboration platform server. The collaboration platform server forwards the invitation to Agent B. Upon receiving the invitation, Agent B can accept, modify, or reject it. If Agent B proposes modifications (such as adjusting the budget or milestones), Agent A and Agent B will communicate through a negotiation channel in multiple rounds until both parties reach an agreement on the collaboration terms.

[0144] If the two parties cannot reach an agreement during the negotiation process, the negotiation fails and the cooperation is cancelled.

[0145] After both parties reach an agreement on the collaboration terms, the collaboration platform server generates a collaboration contract based on the negotiated terms. The collaboration platform server sends the contract to both Agent A and Agent B, and both parties sign the contract using their respective private keys and submit it to the platform. After the platform verifies the validity of both parties' signatures, it stores the contract and both parties' signatures together, and the contract becomes effective. The platform returns the contract address to Agent A and Agent B respectively for subsequent verification.

[0146] After the contract takes effect, agent A and agent B enter the task execution phase. They exchange data or intermediate results during task execution through an encrypted messaging channel provided by the collaboration platform server. The platform records the execution logs of their interactions for subsequent traceability.

[0147] Upon completion of the task, Agent A verifies the deliverables submitted by Agent B. If the verification is successful, Agent A sends an acceptance confirmation instruction to the collaboration platform server. Responding to this instruction, the collaboration platform server, according to the payment terms in the collaboration contract, uses the payment adapter to invoke an external payment system to transfer the agreed-upon amount of tokens from Agent A's account to Agent B's account. After completing the transfer, the external payment system returns a payment success notification to the collaboration platform server, which then sends the payment completion notification to both Agent A and Agent B.

[0148] After payment is completed, agents A and B submit their mutual trust scores to the collaboration platform server. Upon receiving the submitted scores, the collaboration platform server stores the scores in the trust score table and triggers an incremental update to the reputation graph—adding or updating the corresponding directed edge weights in the memory graph for subsequent matching engine calculations of trust transfer scores.

[0149] To facilitate better implementation of the capability-matching-based multi-agent cooperative interaction method provided in the embodiments of this application, the embodiments of this application also provide a capability-matching-based multi-agent cooperative interaction device based on the above-described capability-matching-based multi-agent cooperative interaction method. The meanings of the terms used are the same as in the above-described capability-matching-based multi-agent cooperative interaction method, and specific implementation details can be found in the descriptions in the method embodiments.

[0150] Please see Figure 10 , Figure 10 This application provides a structural block diagram of a multi-agent collaborative interaction device based on capability matching. The device includes: a first receiving unit 301, used to receive a collaboration request published by a first agent; a first acquiring unit 302, used to acquire capability information of multiple candidate agents, and based on the collaboration request and the capability information of each candidate agent, determine the capability fit of each candidate agent relative to the collaboration request, the capability gap that can compensate for the collaboration request, and the trustworthiness between the first agent and the candidate agents; a first determining unit 303, used to determine the comprehensive matching score corresponding to each candidate agent based on the capability fit, capability gap, and trustworthiness; a filtering unit 304, used to filter out a second agent that fits the collaboration request from among the candidate agents based on the comprehensive matching score, and push the information of the second agent to the first agent; and an execution unit 305, used to respond to the first agent initiating a collaboration invitation to the second agent, establish a collaboration relationship between the first agent and the second agent, and execute the collaboration task corresponding to the collaboration request based on the collaboration relationship.

[0151] In some embodiments, the first acquisition unit 302 may include: an extraction subunit, configured to extract features from the collaboration requirements to obtain a requirement feature vector, and extract features from the capability information of each candidate agent to obtain a capability feature vector, and calculate the similarity between the requirement feature vector and the capability feature vector as the capability fit; a matching subunit, configured to extract the set of skill requirements from the collaboration requirements, obtain the set of skill tags for each candidate agent, match the set of skill requirements with the set of skill tags, and determine the capability gap degree based on the matching result; and a first calculation subunit, configured to obtain a pre-constructed reputation graph, calculate the trust transfer score between the first agent and the candidate agents based on the reputation graph, and determine the credibility based on the trust transfer score.

[0152] In some embodiments, the apparatus may further include: a retrieval unit, configured to retrieve, from a database of candidate agent capability feature vectors, multiple candidate agents whose matching degree with the requirement feature vectors meets preset conditions.

[0153] In some embodiments, the filtering unit 304 may include: a first acquisition subunit, configured to acquire the filtering constraints in the collaboration requirements and filter each candidate agent according to the filtering constraints; and a filtering subunit, configured to select the candidate agents whose comprehensive matching scores meet the preset requirements from the filtered candidate agents as the second agent.

[0154] In some embodiments, the apparatus may further include: a construction unit, configured to construct a directed weighted graph using decentralized identity identifiers of multiple agents as nodes and historical mutual rating scores between agents as edge weights, to obtain a reputation graph.

[0155] In some embodiments, the apparatus may further include: a second receiving unit, configured to receive mutual score scores in a collaborative completion event and update edge weights in the reputation graph based on the mutual score scores.

[0156] In some embodiments, the first computational subunit may be specifically used to: perform random walk iterative computation in the reputation graph with the first agent as the root node until the convergence condition is met, and determine the trust transfer score between the first agent and the candidate agents based on the iterative computation result.

[0157] In some embodiments, the execution unit 305 may include: an establishment subunit, configured to establish a negotiation channel between the first and second intelligent agents in response to a first intelligent agent sending a collaboration invitation to a second intelligent agent, for the first and / or second intelligent agents to negotiate collaboration terms; a generation subunit, configured to generate a collaboration contract based on the negotiated collaboration terms in response to the first and second intelligent agents reaching an agreement; a first receiving subunit, configured to receive a first signature of the collaboration contract by the first intelligent agent based on a first private key and a second signature of the collaboration contract by the second intelligent agent based on a second private key, and store the collaboration contract, the first signature, and the second signature; a recording subunit, configured to provide an encrypted message channel for the first and second intelligent agents during the execution of the collaboration task and record an execution log; and a transfer subunit, configured to trigger a payment operation based on the payment conditions in the collaboration contract in response to an acceptance confirmation instruction from the first intelligent agent, transferring resources from the account of the first intelligent agent to the account of the second intelligent agent.

[0158] In some embodiments, the device may further include: a third receiving unit, configured to, in response to a registration request from an intelligent agent, receive a public key submitted by the intelligent agent, generate a decentralized identity identifier for the intelligent agent based on the public key, and store the correspondence between the decentralized identity identifier and the public key, thereby completing the registration of the intelligent agent; wherein the intelligent agent includes at least a first intelligent agent and a second intelligent agent.

[0159] In some embodiments, the first receiving unit 301 may include: a second receiving subunit, configured to receive a collaboration request, the collaboration request carrying a decentralized identity identifier of a first intelligent agent and a first digital signature generated based on the private key of the first intelligent agent; a first verification subunit, configured to query the corresponding public key based on the decentralized identity identifier of the first intelligent agent and verify the validity of the first digital signature using the public key; and a publishing subunit, configured to publish the collaboration request if the verification is successful.

[0160] In some embodiments, the device may further include: a fourth receiving unit, configured to receive a capability profile submitted by an intelligent agent, the capability profile carrying the intelligent agent's decentralized identity identifier and a digital signature generated based on the intelligent agent's private key; a verification unit, configured to query the corresponding public key based on the decentralized identity identifier and verify the validity of the digital signature using the public key; a verification unit, configured to verify the capability profile if the verification passes; and a storage unit, configured to store the capability profile in a decentralized storage network after the verification passes, obtain the content address corresponding to the capability profile, and store the correspondence between the decentralized identity identifier and the content address; wherein the intelligent agent includes a first intelligent agent and a second intelligent agent.

[0161] In some embodiments, the storage unit may include: a second computing subunit, used to calculate the content hash value of the capability file, upload the capability file to the decentralized storage network, and receive the content address returned by the decentralized storage network.

[0162] In some embodiments, the apparatus may further include: an update unit for extracting skill tags from the capability profile and updating the inverted index based on the skill tags; an extraction unit for extracting features from the capability profile to obtain capability feature vectors and storing the capability feature vectors in a capability feature vector library; and a submission unit for submitting the content address and / or content hash value to the blockchain for notarization.

[0163] In some embodiments, the verification unit may include: a second acquisition subunit, used to acquire the test dataset and verification criteria associated with the capability type corresponding to the capability profile; a third receiving subunit, used to receive a zero-knowledge proof submitted by the agent, wherein the zero-knowledge proof is generated locally by the agent based on the test dataset and the agent's model, and is used to prove that the model's test results on the test dataset meet the verification criteria; a second verification subunit, used to call the verifier to verify the zero-knowledge proof; and an adding subunit, used to add a verified mark to the capability profile and store the proof address of the zero-knowledge proof if the verification passes.

[0164] This application discloses a multi-agent collaborative interaction device based on capability matching. A first receiving unit 301 receives a collaboration request issued by a first agent; a first acquiring unit 302 acquires capability information of multiple candidate agents, and based on the collaboration request and the capability information of each candidate agent, determines the capability fit of each candidate agent relative to the collaboration request, the capability gap that can compensate for the collaboration request, and the trustworthiness between the first agent and the candidate agents; a first determining unit 303 determines the comprehensive matching score corresponding to each candidate agent based on the capability fit, capability gap, and trustworthiness; a filtering unit 304 filters out a second agent that matches the collaboration request from among the candidate agents based on the comprehensive matching score, and pushes the information of the second agent to the first agent; an execution unit 305 responds to the collaboration invitation initiated by the first agent to the second agent, establishes a collaboration relationship between the first agent and the second agent, and executes the collaboration task corresponding to the collaboration request based on the collaboration relationship. This achieves fully automated interaction from request matching to collaboration completion in a multi-agent system, improving the accuracy of collaboration matching and task execution efficiency.

[0165] Accordingly, embodiments of this application also provide an electronic device. For example... Figure 11 As shown, Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device 400 includes a processor 401 with one or more processing cores, a memory 402 with one or more computer-readable storage media, and a computer program stored in the memory 402 and executable on the processor. The processor 401 and the memory 402 are electrically connected. Those skilled in the art will understand that... Figure 11 The electronic device structures shown herein do not constitute a limitation on electronic devices and may include, but are not limited to, those shown. Figure 11 It can show more or fewer parts, or combine certain parts, or arrange different parts.

[0166] The processor 401 is the control center of the electronic device 400. It connects various parts of the electronic device 400 through various interfaces and lines. By running or loading software programs and / or modules stored in the memory 402, and calling data stored in the memory 402, it performs various functions of the electronic device 400 and processes data, thereby monitoring the electronic device 400 as a whole.

[0167] In this embodiment, the processor 401 in the electronic device 400 loads the instructions corresponding to the processes of one or more applications into the memory 402 according to the following steps, and the processor 401 runs the applications stored in the memory 402 to realize various functions: The system receives a collaboration request from a first agent; acquires capability information from multiple candidate agents; and, based on the collaboration request and the capability information of each candidate agent, determines the capability fit of each candidate agent relative to the collaboration request, the capability gap that can be filled in the collaboration request, and the trustworthiness between the first agent and the candidate agents; determines the comprehensive matching score for each candidate agent based on the capability fit, capability gap, and trustworthiness; selects a second agent that fits the collaboration request from among the candidate agents based on the comprehensive matching score, and pushes the information of the second agent to the first agent; responds to the first agent's invitation to the second agent to initiate a collaboration relationship, establishes a collaboration relationship between the first agent and the second agent, and executes the collaboration task corresponding to the collaboration request based on the collaboration relationship.

[0168] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0169] Optional, such as Figure 11 As shown, the electronic device 400 may further include a display 403 and an input unit 404. The processor 401 is electrically connected to both the display 403 and the input unit 404. Those skilled in the art will understand that... Figure 11 The electronic device structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0170] Display 403 can be used to display a graphical user interface (GUI) and receive operation commands generated by the user interacting with the GUI. Display 403 may include a display panel and a touch panel. The display panel can be used to display information input by the user or information provided to the user, as well as various graphical user interfaces of the electronic device. These graphical user interfaces can be composed of graphics, guidance information, icons, video, and any combination thereof. Optionally, the display panel can be configured using a liquid crystal display (LCD), organic light-emitting diode (OLED), or other similar technologies. The touch panel can be used to collect touch operations performed by the user on or near it (such as operations performed by the user using a finger, stylus, or any suitable object or accessory on or near the touch panel), generate corresponding operation commands, and execute the corresponding program according to the operation commands. Optionally, the touch panel may include a touch detection device and a touch controller.

[0171] The touch detection device detects the user's touch location and the signal generated by the touch operation, transmitting the signal to the touch controller. The touch controller receives touch information from the touch detection device, converts it into touch point coordinates, and sends it to the processor 401. It can also receive and execute commands from the processor 401. The touch panel can cover the display panel. When the touch panel detects a touch operation on or near it, it transmits the information to the processor 401 to determine the type of touch event. Subsequently, the processor 401 provides corresponding visual output on the display panel based on the type of touch event. In this embodiment, the touch panel and display panel can be integrated into the display 403 to achieve input and output functions. However, in some embodiments, the touch panel and display panel can be implemented as two independent components to achieve input and output functions. That is, the display 403 can also be used as part of the input unit 404 to achieve input functions.

[0172] The input unit 404 can be used to receive input numbers, characters, or user characteristic information (such as fingerprints, iris, facial information, etc.), and generate keyboard, mouse, joystick, optical, or trackball signal inputs related to user settings and function control.

[0173] In some embodiments, the electronic device may further include an audio circuit, which can provide an audio interface between the user and the electronic device via a speaker and a microphone. The audio circuit can convert received audio data into electrical signals and transmit them to the speaker, where the speaker converts them into sound signals for output. Conversely, the microphone converts collected sound signals into electrical signals, which are then received by the audio circuit, converted back into audio data, and output to processor 401 for processing. The audio data is then transmitted via radio frequency circuitry to, for example, another electronic device, or output to memory 402 for further processing. The audio circuit may also include an earphone jack to provide communication between peripheral headphones and the electronic device.

[0174] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0175] To this end, embodiments of this application provide a computer-readable storage medium storing a plurality of computer programs that can be loaded by a digital signal processor to execute steps in any of the capability-matching-based multi-agent cooperative interaction methods provided in embodiments of this application.

[0176] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0177] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0178] Since the computer program stored in the computer-readable storage medium can execute the steps of any of the capability-matching multi-agent cooperative interaction methods provided in the embodiments of this application, the beneficial effects that any of the capability-matching multi-agent cooperative interaction methods provided in the embodiments of this application can achieve can be realized, as detailed in the preceding embodiments, and will not be repeated here.

[0179] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application shall still fall within the scope of this application.< / hash> < / hash> < / hash>

Claims

1. A multi-agent collaborative interaction method based on capability matching, applied to a collaborative platform server, characterized in that, The method includes: Receive collaboration requests from the first intelligent agent; The capability information of multiple candidate intelligent agents is obtained. Based on the collaboration requirements and the capability information of each candidate intelligent agent, the capability adaptability of each candidate intelligent agent relative to the collaboration requirements, the capability gap that can make up for the collaboration requirements, and the trustworthiness between the first intelligent agent and the candidate intelligent agents are determined respectively. Based on the capability fit, the capability gap, and the credibility, a comprehensive matching score is determined for each candidate agent. Based on the comprehensive matching score, a second intelligent agent that is suitable for the collaboration requirements is selected from the candidate intelligent agents, and the information of the second intelligent agent is pushed to the first intelligent agent; In response to the first agent initiating a collaboration invitation to the second agent, a collaboration relationship is established between the first agent and the second agent, and the collaboration task corresponding to the collaboration requirement is executed based on the collaboration relationship.

2. The method according to claim 1, characterized in that, The step of determining, based on the collaboration requirements and the capability information of each candidate agent, the capability fit of each candidate agent relative to the collaboration requirements, the capability gap that can compensate for the collaboration requirements, and the trustworthiness between the first agent and the candidate agents, includes: Feature extraction is performed on the collaboration requirements to obtain a requirement feature vector, and feature extraction is performed on the capability information of each candidate agent to obtain a capability feature vector. The similarity between the requirement feature vector and the capability feature vector is calculated as the capability fit. Extract the set of skill requirements from the collaboration requirements, obtain the set of skill tags for each candidate agent, match the set of skill requirements with the set of skill tags, and determine the capability gap based on the matching result; Obtain a pre-constructed reputation graph, calculate the trust transfer score between the first agent and the candidate agent based on the reputation graph, and determine the credibility based on the trust transfer score.

3. The method according to claim 2, characterized in that, The method further includes: In the candidate agent capability feature vector library, retrieve the multiple candidate agents whose matching degree with the requirement feature vector meets the preset conditions; The step of selecting a second agent that matches the collaboration requirements from among the candidate agents based on the comprehensive matching score includes: Obtain the screening constraints in the collaboration requirements, and filter the candidate agents according to the screening constraints; From the filtered candidate agents, the candidate agents whose comprehensive matching scores meet the preset requirements are selected as the second agent.

4. The method according to claim 2, characterized in that, The method further includes: Using decentralized identity identifiers of multiple intelligent agents as nodes and historical mutual rating scores between intelligent agents as edge weights, a directed weighted graph is constructed to obtain the reputation graph. The method further includes: Receive the mutual score in the collaboration completion event, and update the edge weights in the reputation graph based on the mutual score; The step of calculating the trust transfer score between the first agent and the candidate agents based on the reputation graph includes: Using the first agent as the root node, perform random walk iterative calculations in the reputation graph until the convergence condition is met, and determine the trust transfer score between the first agent and the candidate agents based on the iterative calculation results.

5. The method according to claim 1, characterized in that, The step of responding to the first agent initiating a collaboration invitation to the second agent, establishing a collaboration relationship between the first agent and the second agent, and executing the collaboration task corresponding to the collaboration requirement based on the collaboration relationship includes: In response to the first intelligent agent sending the cooperation invitation to the second intelligent agent, a negotiation channel is established between the first intelligent agent and the second intelligent agent to allow the first intelligent agent and / or the second intelligent agent to negotiate the cooperation terms; In response to the first intelligent agent and the second intelligent agent reaching an agreement on the negotiated cooperation terms, a cooperation contract is generated according to the negotiated cooperation terms. Receive the first signature of the first smart agent on the cooperation contract based on the first private key and the second signature of the second smart agent on the cooperation contract based on the second private key, and store the cooperation contract, the first signature and the second signature; During the execution of the collaborative task, an encrypted message channel is provided for the first and second intelligent agents, and an execution log is recorded; In response to the acceptance confirmation instruction from the first intelligent agent, a payment operation is triggered according to the payment conditions in the collaboration contract, transferring resources from the account of the first intelligent agent to the account of the second intelligent agent.

6. The method according to claim 1, characterized in that, Before receiving the collaboration request issued by the first intelligent agent, the method also includes: In response to the registration request of the intelligent agent, the system receives the public key submitted by the intelligent agent, generates a decentralized identity identifier for the intelligent agent based on the public key, stores the correspondence between the decentralized identity identifier and the public key, and completes the registration of the intelligent agent. The intelligent agent includes at least the first intelligent agent and the second intelligent agent.

7. The method according to claim 6, characterized in that, The receiving of the collaboration request issued by the first intelligent agent includes: The collaboration request is received, and the collaboration request carries the decentralized identity of the first intelligent agent and a first digital signature generated based on the private key of the first intelligent agent; The public key is queried based on the decentralized identity of the first intelligent agent, and the validity of the first digital signature is verified using the public key. If the verification passes, the collaboration request will be published.

8. The method according to claim 1, characterized in that, Before receiving the collaboration request issued by the first intelligent agent, the method also includes: Receive capability profiles submitted by intelligent agents, the capability profiles carrying the decentralized identity of the intelligent agents and a digital signature generated based on the private key of the intelligent agents; The public key corresponding to the decentralized identity is retrieved, and the validity of the digital signature is verified using the public key. If the verification passes, the capability profile will be validated. After verification, the capability profile is stored in a decentralized storage network, the content address corresponding to the capability profile is obtained, and the correspondence between the decentralized identity and the content address is stored. The intelligent agent includes the first intelligent agent and the second intelligent agent.

9. The method according to claim 8, characterized in that, The step of storing the capability profile in a decentralized storage network and obtaining the content address corresponding to the capability profile includes: Calculate the content hash value of the capability file, upload the capability file to the decentralized storage network, and receive the content address returned by the decentralized storage network; The method further includes: Extract skill tags from the ability profile and update the inverted index based on the skill tags; The capability profile is subjected to feature extraction to obtain capability feature vectors, and the capability feature vectors are stored in the capability feature vector library; Submit the content address and / or the content hash value to the blockchain for evidence storage.

10. The method according to claim 8, characterized in that, The verification of the capability profile includes: Obtain the test dataset and verification standards associated with the capability type corresponding to the capability profile; The agent receives a zero-knowledge proof submitted by the agent, which is generated locally by the agent based on the test dataset and the agent's model, and is used to prove that the test results of the model on the test dataset meet the verification criteria. Call the verifier to verify the zero-knowledge proof; If the verification passes, a verified mark is added to the capability profile, and the proof address of the zero-knowledge proof is stored.

11. A multi-agent collaborative interaction device based on capability matching, applied to a collaborative platform server, characterized in that, The device includes: The first receiving unit is used to receive the collaboration request issued by the first intelligent agent; The first acquisition unit is used to acquire the capability information of multiple candidate intelligent agents, and based on the collaboration requirements and the capability information of each candidate intelligent agent, determine the capability adaptability of each candidate intelligent agent relative to the collaboration requirements, the capability gap that can make up for the collaboration requirements, and the trustworthiness between the first intelligent agent and the candidate intelligent agents respectively. The first determining unit is used to determine the comprehensive matching score corresponding to each candidate agent based on the capability adaptability, the capability gap, and the credibility. The filtering unit is used to filter out a second intelligent agent that is suitable for the collaboration requirements from the candidate intelligent agents based on the comprehensive matching score, and push the information of the second intelligent agent to the first intelligent agent; An execution unit is configured to respond to a collaboration invitation initiated by the first agent to the second agent, establish a collaboration relationship between the first agent and the second agent, and execute the collaboration task corresponding to the collaboration requirement based on the collaboration relationship.

12. An electronic device, characterized in that, The electronic device includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the program to implement the capability-matching-based multi-agent cooperative interaction method as described in any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions adapted for loading by a processor to execute the capability-matching-based multi-agent cooperative interaction method according to any one of claims 1 to 10.

Images