Multi-agent communication method based on ontology semantic alignment, equipment and storage medium

Abstract

Embodiments of the present disclosure disclose a multi-agent communication method based on ontology semantic alignment, equipment and storage medium. A specific embodiment of the method comprises: performing communication task information extraction on conversation information to obtain a communication task information set; determining a joint semantic drift degree set; generating a communication concept mapping information set; performing symbol consistency verification processing on the communication concept mapping information set to obtain a consistency verification result set; storing the communication concept mapping information set that passes the characterization verification into a dynamic semantic consensus database; performing semantic constraint conversion on the conversation information to obtain target sending conversation information; and sending the target sending conversation information to a receiving agent for receiving and multi-round conversation communication by the receiving agent. The embodiment is suitable for real-time multi-round communication of multi-agents, can improve the accuracy and efficiency of multi-agent communication cooperation, reduce the waste of communication transmission resources, and shorten the communication cooperation time.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the field of computer technology, and more specifically to a multi-agent communication method, apparatus, and storage medium based on ontology semantic alignment. Background Technology

[0002] In multi-agent systems, agents are typically driven by different cue words, role settings, toolchains, and domain knowledge to perceive the environment and make decisions. However, in multi-turn dialogues, multiple agents may use the same natural language vocabulary, but the logical referents of concepts within each agent may differ, leading to semantic drift and context illusion. This necessitates multiple rounds of communication among the agents for clarification, resulting in a significant waste of communication resources. For multi-agent communication based on ontology semantic alignment, the common approach is to communicate between agents through fixed communication structure protocols or static ontology.

[0003] However, in practice, it has been found that when using the above methods to communicate among multiple agents based on ontology semantic alignment, the following technical problems often arise: Using fixed communication structure protocols or static ontology for multi-agent communication makes it difficult to handle new concepts and attributes that may temporarily emerge during multi-round real-time communication in multi-agent systems, and structural mismatches are prone to occur in cross-domain multi-agent communication collaboration; furthermore, all communication information between multiple agents requires high-cost negotiation and repeated rounds of communication to resolve concept conflicts, leading to increased consumption of communication transmission resources, reduced accuracy and efficiency of multi-agent communication, and prolonged communication collaboration time.

[0004] The information disclosed in this background section is only intended to enhance the understanding of the background of the present disclosure concept, and therefore may contain information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0006] Some embodiments of this disclosure propose a multi-agent communication method, apparatus, and storage medium based on ontology semantic alignment to solve one or more of the technical problems mentioned in the background section above.

[0007] In a first aspect, some embodiments of this disclosure provide a multi-agent communication method based on ontology semantic alignment, comprising: extracting communication task information from the session information to be sent by the sending agent in a multi-agent communication system to obtain a communication task information set, wherein the multi-agent communication system further comprises: a receiving agent and a dynamic semantic consensus database; determining a joint semantic drift set of the communication task information set, the receiving agent's local ontology knowledge base, and the current task context information; in response to determining that at least one joint semantic drift satisfies a preset condition in the joint semantic drift set, using a semantic mapping large language model, based on the at least one joint semantic drift... The communication concept mapping information set is generated by taking the communication task information set corresponding to the degree, the aforementioned local ontology knowledge base, and the aforementioned current task context information; the symbol consistency verification processing is performed on the aforementioned communication concept mapping information set to obtain a consistency verification result set; the communication concept mapping information set corresponding to at least one consistency verification result that passes the representation verification is stored in the dynamic semantic consensus database; based on the aforementioned dynamic semantic consensus database and the communication task information set that does not meet the aforementioned preset conditions, the aforementioned session information is semantically constrained to obtain target sending session information; the aforementioned target sending session information is sent to the aforementioned receiving agent for the receiving agent to receive and conduct multi-round session communication.

[0008] Secondly, some embodiments of this disclosure provide a multi-agent communication device based on ontology semantic alignment, comprising: an information extraction unit configured to extract communication task information from the session information to be sent by the sending agent in the multi-agent communication system to obtain a communication task information set, wherein the multi-agent communication system further comprises: a receiving agent and a dynamic semantic consensus database; a determination unit configured to determine a joint semantic drift set of the communication task information set, the receiving agent's local ontology knowledge base, and the current task context information; and a generation unit configured to, in response to determining that at least one joint semantic drift satisfying a preset condition exists in the joint semantic drift set, utilize a semantic mapping large language model to generate a communication task corresponding to the at least one joint semantic drift. The system comprises: an information set, the aforementioned local ontology knowledge base, and the aforementioned current task context information; a symbolic consistency verification unit configured to perform symbolic consistency verification on the aforementioned communication concept mapping information set to obtain a consistency verification result set; a storage unit configured to store the communication concept mapping information set corresponding to at least one consistency verification result that has passed the characterization verification into a dynamic semantic consensus database; a semantic constraint transformation unit configured to perform semantic constraint transformation on the aforementioned session information based on the aforementioned dynamic semantic consensus database and the communication task information set that does not meet the aforementioned preset conditions to obtain target sending session information; and a sending unit configured to send the aforementioned target sending session information to the aforementioned receiving agent for the receiving agent to receive and conduct multi-round session communication.

[0009] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, such that when the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any implementation of the first aspect.

[0010] Fourthly, some embodiments of this disclosure provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method as described in any implementation of the first aspect.

[0011] The above embodiments of this disclosure have the following beneficial effects: The multi-agent communication method based on ontology semantic alignment in some embodiments of this disclosure is suitable for real-time, multi-round communication between multiple agents, improving the accuracy and efficiency of multi-agent communication collaboration, reducing the waste of communication transmission resources, and shortening the communication collaboration time. Specifically, the reasons why it is difficult to adapt to multi-round real-time scenarios, increasing the consumption of communication transmission resources, reducing the accuracy and efficiency of multi-agent communication, and prolonging the communication collaboration time are as follows: Multi-agent communication using fixed communication structure protocols or static ontology is difficult to handle temporarily emerging new concepts and attributes in multi-round real-time communication within a multi-agent system, and structural mismatch problems are prone to occur in cross-domain multi-agent communication collaboration; simultaneously, all communication information between multiple agents requires high-cost negotiation and repeated multi-round communication negotiations regarding concept conflicts, leading to increased consumption of communication transmission resources, reduced accuracy and efficiency of multi-agent communication, and prolonged communication collaboration time. Based on this, some embodiments of the multi-agent communication method based on ontology semantic alignment disclosed herein can first extract communication task information from the session information to be sent by the sending agent in the multi-agent communication system to obtain a communication task information set. The multi-agent communication system further includes a receiving agent and a dynamic semantic consensus database. Here, by extracting information before sending the session information, semantic drift suppression can be brought forward to the entry point of the execution chain, avoiding further expansion of erroneous semantic information. Secondly, a joint semantic drift degree set is determined from the aforementioned communication task information set, the receiving agent's local ontology knowledge base, and the current task context information. Determining the joint semantic drift degree set facilitates the subsequent execution of different processing methods, avoiding high-cost negotiation communication for all session information, balancing communication accuracy and real-time performance, and reducing the overhead of communication and computing resources. Thirdly, in response to determining that at least one joint semantic drift degree satisfying a preset condition exists in the aforementioned joint semantic drift degree set, a communication concept mapping information set is generated using a semantic mapping large language model based on the communication task information set corresponding to the at least one joint semantic drift degree, the receiving local ontology knowledge base, and the current task context information. Here, a communication concept mapping information set is generated from the communication task information set that meets preset conditions. A hierarchical control strategy of lightweight detection and on-demand semantic alignment is employed, along with the use of a semantic mapping large language model to improve the accuracy of the communication concept mapping information set. Next, the communication concept mapping information set undergoes symbolic consistency verification to obtain a consistency verification result set. This symbolic consistency verification process reduces the illusionary alignment generated by the semantic mapping large language model, combining the flexibility of a neural model with the reliability of formal logic. Subsequently, the communication concept mapping information set corresponding to at least one consistency verification result that passes the representation verification is stored in a dynamic semantic consensus database.Here, the dynamic semantic consensus database can be used to store verified communication concept mapping information, enabling the reuse of the same semantics in multi-round communication, reducing redundant negotiation communication between multiple agents, reducing communication resource consumption and the time required for subsequent semantic transformation, and improving the efficiency of multi-agent communication. Then, based on the dynamic semantic consensus database and the communication task information set that does not meet the above preset conditions, the above session information is semantically constrained and transformed to obtain the target sending session information. Here, the efficiency of semantic constraint transformation and the quality of the target sending session information can be improved, flexibly adapting to the emergence of new concepts and attributes during communication, and effectively reducing the probability of semantic offset problems. Finally, the above target sending session information is sent to the above receiving agent for reception and multi-round session communication. Here, the waste of communication transmission resources can be reduced, communication efficiency and real-time performance can be improved, and the duration of multi-agent communication collaboration can be reduced. Therefore, the multi-agent communication method based on ontology semantic alignment is suitable for real-time multi-round communication among multiple agents. It can promptly detect semantic deviations in the communication process, trigger semantic alignment processing of communication concept mapping and consistency verification of alignment results as needed, and enable agents to cache and reuse alignment results in multi-round communication while maintaining the independence of their respective ontology knowledge bases. This can improve the accuracy and efficiency of multi-agent communication collaboration, reduce the waste of communication transmission resources, and shorten the communication collaboration time. Attached Figure Description

[0012] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0013] Figure 1 This is a flowchart of some embodiments of the multi-agent communication method based on ontology semantic alignment according to the present disclosure; Figure 2 This is a schematic diagram of the structure of some embodiments of a multi-agent communication device based on ontology semantic alignment according to the present disclosure; Figure 3 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0014] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0015] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0016] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0017] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0018] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0019] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] Figure 1 A flow 100 of some embodiments of a multi-agent communication method based on ontology semantic alignment according to this disclosure is shown. This multi-agent communication method based on ontology semantic alignment includes the following steps: Step 101: Extract communication task information from the session information to be sent by the sending agent in the multi-agent communication system to obtain a communication task information set.

[0021] In some embodiments, the executing entity (e.g., an electronic device) of the above-described ontology-based semantic alignment multi-agent communication method can extract communication task information from the session information to be sent by the sending agent in the multi-agent communication system to obtain a communication task information set. The multi-agent communication system further includes a receiving agent and a dynamic semantic consensus database. The multi-agent communication system can be a system that coordinates and controls multiple agents to make decisions in cross-domain (e.g., cross-domain supply chain collaboration, medical collaboration) collaborative communication. The receiving agent can be an agent used to receive session information. The dynamic semantic consensus database can be a database used to store and reuse ontologies that collaboratively confirm the absence of semantic drift between agents. The sending agent can be an agent used to send session information. The sending agent can be software, hardware, or a system. The session information can be information used for communication between multiple agents. The communication task information in the communication task information set can include, but is not limited to, at least one of the following: concepts, actions, attributes, values, units, times, and quantities representing the semantic information of the session information. The concepts can be entities, objects, categories, and things in the session information. The aforementioned actions can be behaviors, operations, relationships, and predicates in the aforementioned session information. For example, if the aforementioned session information is "increase the sampling frequency of the temperature sensor to 100Hz", then the key concept can be "temperature sensor", the action can be "increase", the attribute can be "sampling frequency", and the value can be "100Hz".

[0022] In some optional implementations of certain embodiments, the above-described extraction of communication task information from the session information to be sent by the sending agent in a multi-agent communication system to obtain a communication task information set may include the following steps: The first step is to perform text preprocessing on the above conversation information to obtain preprocessed conversation information.

[0023] The second step is to perform word segmentation on the preprocessed session information to obtain a communication word segmentation set.

[0024] The third step involves performing dependency parsing on the aforementioned communication word segmentation set to obtain a conversation dependency parsing tree. This conversation dependency parsing tree can be a tree structure with communication word segmentation sets as nodes and the relationships between them as connecting edges. These relationships can include, but are not limited to, at least one of the following: subject-verb, verb-object, indirect object, and attributive-head.

[0025] The fourth step involves performing rule matching to extract the initial concept information set from the aforementioned session dependency syntax tree. This initial concept information set can include key concepts, actions, attributes, and values ​​from the aforementioned session information. In practice, the executing entity can use regular expressions to extract nouns, predicate cores, modifiers, and dependencies from the aforementioned session dependency syntax tree to obtain the initial concept information set.

[0026] The fifth step involves coreference resolution of the initial conceptual information set to obtain the communication task information set. In practice, the executing entity can utilize the TruthfulRAG (Truthful Retrieval-Augmented Generation) model to resolve coreference in the initial conceptual information set and obtain the communication task information set.

[0027] Step 102: Determine the joint semantic drift set of the communication task information set, the receiving agent's local ontology knowledge base, and the current task context information.

[0028] In some embodiments, the executing entity may determine a joint semantic drift set of the communication task information set, the receiving agent's local ontology knowledge base, and the current task context information. The receiving agent may be an agent used to receive session information. The receiving local ontology knowledge base may be a database owned by the receiving agent itself for storing knowledge structures describing concepts, relationships, and constraints. The current task context information may be a data stream that changes in real time as communication rounds / times progress. The current task context information may include: task objective information, historical interaction trajectory information, environmental state information of multi-agent operation, and shared context. The task objective information may be a specific task in the session information (e.g., writing a Python script), rules to be followed (e.g., the value range of each parameter in the Python script), and the current state information of task execution (e.g., requirements analysis completed, code being coded). The historical interaction trajectory information may be dialogue records between the receiving agent and the sending agent, sequences of actions already executed, and feedback information from the operating environment to the action sequences. The environmental state information may be a current snapshot of the external world operated by the agent. For example, the environmental state information may be the current codebase state or error logs. The shared context mentioned above can be information exchanged between agents (e.g., agents' intentions and requests). The joint semantic drift degree in the joint semantic drift degree set mentioned above can characterize the degree of semantic difference between the communication task information set and the received local ontology knowledge base and the current task context information.

[0029] In some optional implementations of certain embodiments, determining the joint semantic drift set of the communication task information set, the receiving agent's local ontology knowledge base, and the current task context information may include the following steps: The first step is to determine the receiving agent's local ontology knowledge base based on the message header information in the aforementioned session information. In practice, the executing entity can extract agent identification information from the message header information in the aforementioned session information. This agent identification information can be the agent's URI (Uniform Resource Identifier). Then, based on the agent identification information, the receiving local ontology knowledge base is determined.

[0030] The second step involves performing multi-granularity matching on the aforementioned communication task information set and the aforementioned local ontology knowledge base to obtain a first matching receiving ontology set. The first matching receiving ontology in this set can be an ontology located in the aforementioned local ontology knowledge base and matching the aforementioned communication task information set, obtained through fuzzy matching and precise matching. The multi-granularity matching can include: precise matching based on complete consistency with the communication task information set, and fuzzy matching based on synonyms and aliases.

[0031] The third step involves performing proximity embedding matching on the aforementioned communication task information set and the aforementioned local ontology knowledge base to obtain a second matching receiving ontology set. The second matching receiving ontology in this set can be an ontology obtained through proximity relationships between semantic vectors, located in the aforementioned local ontology knowledge base, and matching the aforementioned communication task information set. This proximity embedding matching can be performed using an ANN (Approximate Nearest Neighbor) algorithm.

[0032] The fourth step involves determining the multi-dimensional semantic matching degree set of the first matching receiving ontology set, the second matching receiving ontology set, and the communication task information set. The multi-dimensional semantic matching degree can be a value obtained by fusing the confidence of the dependency arcs of the extracted communication task information set with the matching degree (e.g., exact matching degree, synonym matching degree, parent-child relationship matching degree, nearest neighbor embedding similarity) between the first and second matching receiving ontology sets and the communication task information set. This value is used to obtain the semantic deviation degree from the confidence of text analysis and the matching degree of ontology mapping. In practice, the executing entity performs the following determination steps for each communication task information: First, determine the confidence of the dependency arcs of the communication task information. Second, determine the relational path information set of the actions included in the first matching receiving ontology set, the second matching receiving ontology set, and the communication task information. Third, filter out at least one relational path information with a reachable semantic path from the relational path information set. Then, determine at least one first matching receiving ontology and at least one second matching receiving ontology corresponding to at least one relational path information as the target matching receiving ontology set. Subsequently, the concept matching degree set, action matching degree set, and attribute value matching degree set of the target matching receiving ontology set and the communication task information are determined. The concept matching degree in the concept matching degree set can be: exact matching degree, synonym matching degree, parent-child relationship matching degree, and nearest neighbor embedding similarity. Specifically, the exact matching degree can be a value where the concept matching degree is 1 if the target matching receiving ontology and the communication task information are a perfect match. The synonym matching degree can be a value in the range [0.8, 0.9] if the target matching receiving ontology and the communication task information are synonyms. The parent-child relationship matching degree can be a value in the range [0.6, 0.8] if the communication task information is a direct parent or direct child of the target matching receiving ontology. The nearest neighbor embedding similarity can be the cosine similarity between the target matching receiving ontology and the communication task information. The action matching degree in the action matching degree set can be the similarity between the attribute / operation information in the target matching receiving ontology and the actions included in the communication task information. The attribute value matching degree in the above attribute value matching degree set can be the matching degree of whether the attribute belongs to a concept and whether the value conforms to the data type and range constraints of the attribute. For example, the above attribute value matching degree can be 1 if both the attribute and the value conform, 0.5 if they partially conform, and 0 otherwise. Then, the concept matching degree set, action matching degree set, attribute value matching degree set, and confidence degree set are weighted and summed to obtain the multi-dimensional semantic matching degree.

[0033] The fifth step is to determine the set of multidimensional context consistency values ​​for the aforementioned communication task information set and the aforementioned current task context information. These multidimensional context consistency values ​​can be used to measure the degree of coordination between session information and local context, dialogue state, and historical session information. The multidimensional context consistency values ​​can include: task stack alignment, temporal consistency, ontology context constraint consistency, and message header thread consistency. Task stack alignment characterizes whether the key concepts and actions included in the communication task information are consistent with the currently unfinished task objectives. Temporal consistency characterizes whether the values ​​included in the communication task information have logical conflicts with the values ​​included in the historical session information. Ontology context constraint consistency characterizes the consistency constraints determined using constraint information included in the received local ontology knowledge base. Message header thread consistency characterizes whether the URI, type identifier, and message body content in the message header are consistent.

[0034] Step 6: Determine the set of differences between the preset value and the weighted sum of the corresponding multi-dimensional semantic matching degree values ​​in the multi-dimensional semantic matching degree set and the multi-dimensional context consistency degree value in the multi-dimensional context consistency degree value set, as the joint semantic drift degree set. The preset value can be a pre-defined value, and its value can be 1.

[0035] Step 103: In response to determining that there exists at least one joint semantic drift degree that satisfies a preset condition in the joint semantic drift degree set, a communication concept mapping information set is generated using the semantic mapping large language model based on the communication task information set corresponding to at least one joint semantic drift degree, the received local ontology knowledge base, and the current task context information.

[0036] In some embodiments, the executing entity may, in response to determining that at least one joint semantic drift degree in the joint semantic drift degree set satisfies a preset condition, utilize a semantic mapping large language model to generate a communication concept mapping information set based on the communication task information set corresponding to the at least one joint semantic drift degree, the received local ontology knowledge base, and the current task context information. The preset condition may be that the joint semantic drift degree is greater than or equal to a preset offset threshold. The preset offset threshold may be a pre-set value, and its value can be determined according to specific circumstances, without limitation here. The semantic mapping large language model can generate semantic mapping relationships for the input communication task information set corresponding to at least one joint semantic drift degree, the received local ontology knowledge base, and the current task context information, and output a large language model of the communication concept mapping information set. For example, the semantic mapping large language model may be, but is not limited to, at least one of the following: DeepSeek model, ChatGPT (Chat Generative Pre-trained Transformer), and NExT-GPT model. The communication concept mapping information in the aforementioned communication concept mapping information set can be information on the mapping relationship, mapping confidence, attribute correspondence, and attribute transformation rules between the communication task information set and the corresponding ontology in the receiving local ontology knowledge base. The aforementioned ontology can be a knowledge structure used to describe concepts, relationships, and constraints. The aforementioned attribute transformation rules can be attribute transformation functions (e.g., unit conversion functions). The aforementioned mapping relationships can include: subclass mapping relationships and equivalence mapping relationships.

[0037] In some optional implementations of certain embodiments, the above-described generation of a communication concept mapping information set using a semantic mapping large language model, based on the communication task information set corresponding to at least one joint semantic drift degree, the received local ontology knowledge base, and the current task context information, may include the following steps: The first step is to determine the target ontology set in the local ontology knowledge base corresponding to the communication task information set corresponding to at least one joint semantic drift degree, within the local ontology knowledge base of the sending agent. This target ontology set includes a concept set, a relation set, and a logical constraint information set. The target ontology in this target ontology set can be concepts and their neighborhood information obtained through direct matching using retrieval methods such as concept name search and synonym retrieval within the local ontology knowledge base of the sending agent. The neighborhood information can include the parent class, subclass, attributes, relations, and logical constraint information of the concepts. The logical constraint information set can include, but is not limited to, at least one of the following: concept mutual exclusion constraint information, hierarchical inclusion constraint information, attribute type constraint information, attribute value range constraint information, and unit dimension constraint information.

[0038] The second step involves structuring the target ontology set, the corresponding receiving ontology segments in the local receiving ontology knowledge base, and the current task context information to obtain structured input text information. The receiving ontology segments can be directly matched ontology and ontology extension information obtained by searching the target ontology set in the local receiving ontology knowledge base. The ontology extension information can be ontology with one-hop and two-hop relationships to the original ontology. The structured input text information can be text information converted from the target ontology set, the corresponding receiving ontology segments in the local receiving ontology knowledge base, and the current task context information into a data format with a clear format, fields, and semantic relationships. For example, the structured input text information can be JSON formatted text information.

[0039] The third step involves generating zero-shot prompt words based on the structured input text information described above. These zero-shot prompt words can be text instructions used to guide the semantic mapping large language model in processing and generating higher-quality output results. They can also be text instructions employing system instructions, input data, and output format constraints. The system instructions can be instructions used to define the role, task, and output format of the semantic mapping large language model. The input data can be structured input text information. The output format constraints can be instructions that constrain the structured output results of the semantic mapping large language model.

[0040] The fourth step involves inputting the aforementioned zero-shot prompt word information into the semantic mapping large language model for zero-shot inference, resulting in a candidate concept mapping information set. This candidate concept mapping information set includes at least one of the following: mapping type information, mapping confidence, attribute correspondence information, and attribute transformation function. The candidate concept mapping information in this set can be generated by mapping ontologies that have appeared in the conversation information, new concepts that haven't appeared, abbreviations, and cross-domain terms, and can be information about the mapping relationships between the most matching ontologies existing in the local ontology knowledge base. Therefore, compared to existing mapping methods that rely on word similarity, vector nearest neighbor retrieval, fixed label tables, or pre-set template completion, zero-shot inference, obtained through the semantic mapping large language model, does not simply match existing label sets. Instead, it utilizes the semantic mapping large language model to jointly infer the target ontology set, the received ontology segment, and the current task context information without having seen the current specific mapping pair before, outputting concept equivalence mappings, hypernym / hypernym mappings, or attribute transformation mappings, and providing the corresponding attribute transformation function. Therefore, even if new concepts, new attribute combinations, or cross-domain expression differences arise during the communication process, this application can still complete online alignment without relying on manual pre-enumeration of rules.

[0041] The fifth step involves conflict classification processing of the aforementioned candidate concept mapping information set to obtain a communication concept mapping information set. The communication concept mapping information in this set can be obtained by classifying the conflict types included in the candidate concept mapping information. These conflict types can include: concept equivalence relations, concept subordination relations, attribute type conflicts, value range conflicts, and unit / dimensional conflicts. For example, if the candidate concept mapping information maps to a subset or higher-level concept of the target ontology, then the communication concept mapping information is classified as a subordination type. If the candidate concept mapping information maps to a target ontology where the units or dimensions are inconsistent, then the communication concept mapping information is classified as a generation / transformation type. Otherwise, it defaults to the same mapping type. Therefore, conflict classification processing can prevent the semantic mapping large language model from generating inconsistent mapping information with varying granularity and formats, thus improving the accuracy of the communication concept mapping information set.

[0042] In addressing the aforementioned technical problems in the process of adopting technical solutions, the application scenario—cross-domain medical collaborative multi-agent communication—often presents the following technical challenges: Semantic drift arises because each agent uses the same natural language vocabulary across domains, but their internal logical referents differ. Strict rules constrain the use of medical guidelines, clinical norms, and medical abbreviations. Existing large language models suffer from excessively large search spaces, lack of structural constraints and conflict awareness, and poor handling of abbreviations and cross-domain terminology in zero-shot learning. This results in low accuracy of the generated communication concept mapping information, requiring multiple rounds of collaborative clarification, wasting significant communication resources, and prolonging communication time. To meet the following requirements for this application scenario: adaptability to cross-domain multi-agent communication, high precision and accuracy, adaptability to multiple rule constraints and conflict awareness, and adaptability to abbreviations, we have decided to adopt the following solution: In some optional implementations of certain embodiments, the above-described method of utilizing a semantic mapping large language model to generate a communication concept mapping information set based on the communication task information set corresponding to at least one joint semantic drift degree, the received local ontology knowledge base, and the current task context information, and generating target sending session information and sending it to the receiving agent based on the communication concept mapping information set, may include the following steps: The first step involves performing structured parsing on the communication task information set corresponding to at least one joint semantic drift degree to obtain a structured communication information set. This structured communication information set may include: the concepts, parent classes, attributes, and logical constraints of the ontology corresponding to the communication task information, as well as parameter types, units, and format information extracted from the message header. In practice, the executing entity can utilize the Java open-source library OWL API (Application Programming Interface) to extract the structured communication information set from the original concept tags (e.g., URIs, tags), ontology annotations, parent classes, attributes, value ranges, and units corresponding to the communication task information set, as well as the tool parameter declarations and current task context information in the message header.

[0043] The second step involves semantically expanding the target structured information set within the aforementioned communication structured information set to obtain a semantically expanded structured information set. This target structured information set may include at least one of the following: abbreviated text information or cross-domain structured information. Furthermore, the semantically expanded structured information set may include, but is not limited to, at least one of the following: performing zero-shot large language model expansion inference on the tags / annotations of the target structured information to output the most likely complete term and its domain; if it is a multi-semantic term, a predetermined number of candidate expanded concepts and their corresponding confidence levels are pre-set before outputting.

[0044] The third step involves performing a recall retrieval process on the aforementioned semantic extended structured information set, the aforementioned communication structured information set, and the aforementioned local ontology knowledge base to obtain a candidate receiving ontology set. The candidate receiving ontology set can consist of multiple semantic extended structured information vectors that are similar to the semantic extended structured information vectors in terms of both type and attribute. In practice, the executing entity can first use the OWL2Vec* algorithm to determine the corresponding ontology set in the local ontology knowledge base for semantic embedding, obtaining an ontology feature vector set. Then, semantic embedding is performed on multiple communication structured information vectors corresponding to the semantic extended structured information set and the aforementioned communication structured information set to obtain an extended semantic feature vector set. Finally, the entity retrieves the multiple concepts and attributes most similar to the extended semantic feature vector set from the ontology feature vector set, using these as candidate receiving ontology sets.

[0045] The fourth step involves modularly extracting and constructing a candidate receiving ontology subgraph set from the aforementioned candidate receiving ontology set. The candidate receiving ontology subgraphs in this subgraph set can be subgraphs formed by concept hierarchies (inheritance, inclusion, and equivalence classes) of ontology that have one or two hops with the candidate receiving ontology, as well as key constraints and attribute hierarchies such as mutually exclusive classes, domains / ranges, and data types. In practice, the execution entity can first utilize the modularization strategy in the GLaMor graph language model, using the aforementioned candidate receiving ontology set as the center for modular processing to obtain the candidate subgraph set. Then, class hierarchy, attribute hierarchy, and constraint extraction are performed on each candidate subgraph to obtain the candidate receiving ontology subgraph set.

[0046] The fifth step involves extracting triple constraints from the candidate receiving ontology subgraphs to obtain a receiving ontology fragment information set. This receiving ontology fragment information may include concepts, attributes, hierarchical information, and constraint information. In practice, the executing entity can first convert each candidate receiving ontology subgraph into Manchester syntax and then into a set of triples consisting of a subject, relation, and object. Then, for each triple, constraints based on hierarchical inclusion, mutual exclusion, attribute type, and unit / dimension are extracted and classified to obtain the receiving ontology fragment information set.

[0047] Step 6: Construct zero-sample chained constraint prompts for the aforementioned semantically extended structured information set, the aforementioned communication structured information set, the aforementioned received ontology fragment information set, and the aforementioned current task context information to obtain constraint zero-sample prompt information. The constraint zero-sample prompt information can be prompts that explicitly define the task role, list the semantically extended structured information set, the corresponding communication structured information set, the received ontology fragment information set, and the current task context information, constrain the output format, add conflict classification rule information, add prompt text content for step-by-step reasoning of the thought chain, and add zero-sample and structured self-prompting and self-constraining text content. The constraint output format can be a constraint format including the sending ontology, the receiving ontology, including equivalence, subordination, superordinate, transformation (e.g., numerical unit conversion), and incompatible (semantic conflict preventing direct mapping) relation types, confidence level, mapping reasoning basis information, transformation function, and unit conversion description. The zero-sample and structured self-prompting and self-constraining text content can be text content containing multiple constraint detections and conflict correction suggestions. For example, the above text could be: "Please first check the following constraints: Are there mutually exclusive classes? Are the units consistent? Do the attribute types match?", "If a conflict is found, please clearly indicate the conflict type in the output and provide correction suggestions if possible."

[0048] Step 7: Input the aforementioned constraint zero-sample prompt word information into the aforementioned semantic mapping large language model to obtain the communication concept mapping information set, and generate the target sending session information based on the aforementioned communication concept mapping information set and send it to the receiving agent. The implementation method for generating the target sending session information based on the aforementioned communication concept mapping information set and sending it to the receiving agent can refer to the implementation methods given in steps 104-107.

[0049] The above-described technical solution and its related content, as an inventive point of this disclosure, solve the technical problem of "low accuracy of generated communication concept mapping information, wasting a large amount of communication resources, and prolonging communication time." Factors leading to low accuracy of generated communication concept mapping information, wasted communication resources, and prolonged communication time are often as follows: In cross-domain medical collaborative multi-agent collaborative communication scenarios, semantic drift occurs because each agent uses the same natural language vocabulary across domains, but their internal logical referents differ; strict rules constrain the use of medical guidelines, clinical norms, and medical abbreviations; and existing large language models suffer from excessively large search spaces, lack of structural constraints and conflict perception, and poor handling of abbreviations and cross-domain terms in zero-shot learning. This results in low accuracy of generated communication concept mapping information, requiring multiple rounds of collaborative clarification, wasting a large amount of communication resources, and prolonging communication time. Solving these factors can improve the accuracy of generated communication concept mapping information, reduce the waste of communication resources, and shorten communication time. To achieve this effect, this disclosure firstly employs structured parsing and semantic expansion of the target structured information set. Structured parsing fully preserves hierarchical, attribute, and constraint information, avoiding semantic loss issues associated with alignment solely through labels / annotations. Furthermore, zero-sample semantic expansion through context-anchored semantics effectively addresses ambiguities in abbreviated expressions and cross-domain understanding. Secondly, it employs recall retrieval and modular hierarchical extraction construction. Recall retrieval effectively avoids adding the entire local ontology knowledge base to prompts, creating excessively long contexts and diverting the attention of large language models. Modular hierarchical extraction construction compresses the search space while preserving the complete hierarchical structure and associated constraints, improving the accuracy of candidate received ontology subgraphs. Finally, triple constraint extraction effectively addresses the problem of large language models struggling to understand ontology syntax in semantic mapping. Explicitly structuring hard constraints solves the problem of imperceptible implicit constraints, reducing mappings caused by the illusion problem of large language models. Then, zero-shot chained constraint prompts are constructed, bringing hard constraints forward to the semantic mapping large language model generation stage. This effectively reduces the generation of constraint-violation mappings and decreases computational resource usage. Finally, leveraging the semantic understanding and zero-shot reasoning capabilities of the semantic mapping large language model, mappings are generated for new concepts, abbreviations, and cross-domain terms. This effectively alleviates the problems of ontology matching methods relying on labeled data and having poor generalization ability, improves the accuracy of generated communication concept mapping information, reduces communication resource waste, and shortens communication time.

[0050] Step 104: Perform symbol consistency verification on the communication concept mapping information set to obtain a consistency verification result set.

[0051] In some embodiments, the execution entity may perform symbolic consistency verification on the communication concept mapping information set to obtain a consistency verification result set. The consistency verification result in the result set may be a result determining whether the communication concept mapping information set violates inclusion relationships, mutual exclusion relationships, type restrictions, and value range restrictions between concepts.

[0052] In some optional implementations of certain embodiments, the above-described symbol consistency verification process for the communication concept mapping information set to obtain a consistency verification result set may include the following steps: The first step, based on the communication concept mapping information set, is to perform the following input steps: Sub-step 1 involves converting the communication concept mapping information set and the logical constraint information set included in the received local ontology knowledge base into a reasoning format conversion, resulting in a set of formula reasoning triples. The ontology logical constraint relations in the aforementioned ontology logical constraint relation set can be used to describe the logical constraints between ontology entities. The formula reasoning triples in the aforementioned set of formula reasoning triples can be in the form of a subject-predicate-object triple, where the URI identifiers of the ontology corresponding to the communication concept mapping information and the ontology logical constraint information are used as the subject and object, respectively, and the ontology logical constraint information is used as an attribute or relation as the predicate.

[0053] Sub-step 2 involves performing a multi-dimensional logical consistency check on the formula reasoning triple set to obtain a multi-dimensional logical consistency check information set. This multi-dimensional logical consistency check includes at least one of the following: checks to determine whether candidate concept mapping information triggers concept unsatisfiability checks, hierarchical relationship conflicts, attribute domain and value domain mismatch checks, numerical out-of-bounds checks, and unit dimension incompatibility checks. The concept unsatisfiability check can be a check where the concept is false under all possible interpretations. In practice, the executing entity can utilize an OWL (Web Ontology Language) inference engine to perform a multi-dimensional logical consistency check on the formula reasoning triple set to obtain the multi-dimensional logical consistency check information set. The OWL2 inference engine can include, but is not limited to, at least one of the following: HermiT, Pellet, and RDFox.

[0054] Sub-step 3: In response to determining that there is successfully represented verification information in the multidimensional logical consistency verification information set, the successfully represented multidimensional logical consistency verification information is determined as the consistency verification result set.

[0055] The second step involves determining that the multidimensional logical consistency verification information set contains verification information indicating failure. This failure is then subjected to constraint reasoning parsing to obtain a verification parsing information set. The verification parsing information in this set can be a reasoning analysis of the reasons for the verification failure, including the cause of the failure, the type of conflicting constraint, the location of the conflict, the triggering condition, and the suggested correction direction. This constraint reasoning parsing can be performed by inputting the failure-representing multidimensional logical consistency verification information set into a large language model.

[0056] The third step involves regenerating the communication remapping information set based on the aforementioned verification and parsing information set, which serves as the communication concept mapping information set, to repeat the input steps described above. Thus, multi-dimensional logical consistency verification differs from conventional consistency verification in existing technologies, which only checks the legality of the output format, the completeness of fields, or the internal satisfaction of a single ontology. Symbolic consistency verification performs joint verification on the combined result of "communication concept mapping information + hard constraints of the peer ontology + current task context information." It not only checks whether the mapping itself is syntactically valid but also further checks whether the mapping leads to incorrect merging of mutually exclusive concepts, inversion of hierarchical concepts, mismatch of attribute domains or value domains, numerical values ​​exceeding the allowed range, or the absence of acceptable conversion relationships between different units of measurement. If any of these conflicts are found, structured constraint feedback is generated and returned to the semantic mapping large language model for constrained correction, thereby avoiding the situation where only static satisfiability judgments are made, which cannot continuously iterate and correct for real-time dialogue contexts.

[0057] Optionally, the above-mentioned regeneration of the communication remapping information set based on the above-mentioned verification and parsing information set, as the communication concept mapping information set, to perform the above-mentioned input steps again, may include the following steps: The first step is to perform the following verification steps based on each multidimensional logical consistency verification information in the multidimensional logical consistency verification information set corresponding to the above verification parsing information set: Sub-step 1: Determine the number of failed checks performed on the multi-dimensional logical consistency verification information. This number of failed checks can be the current number of consecutive failed checks.

[0058] Sub-step 2 involves performing symbolic consistency verification on the verification parsing information corresponding to the multi-dimensional logical consistency verification information to obtain the target verification information.

[0059] Sub-step 3, in response to determining that the target verification information represents successful verification, stores the communication concept mapping information corresponding to the target verification information in the dynamic semantic consensus database. The aforementioned dynamic semantic consensus database can be a database used to store sets of successfully verified communication concept mapping information.

[0060] Sub-step 4: In response to determining that the target verification information indicates verification failure, and that the number of failed verification executions exceeds a preset execution threshold, the execution of symbol consistency verification ends. The preset execution threshold can be a pre-defined maximum number of consecutive failures. The value of the preset execution threshold can be determined based on specific circumstances and is not limited here.

[0061] Sub-step 5: In response to determining that the target verification information indicates verification failure and that the number of failed verification executions is less than or equal to the preset execution threshold, the verification parsing information, the corresponding target ontology set, the received ontology segment, and the current task context information are input into the semantic mapping large language model to re-map concepts and obtain communication remapping information as a communication concept mapping information set, so as to execute the above input step again.

[0062] In addressing the aforementioned technical challenges in the application scenario—specifically, the generation of detection reports in cross-domain medical collaborative multi-agent communication—often presents the following technical issues: the medical field has numerous compliance requirements; relying solely on symbolic inference engines (e.g., HermiT or Pellet inference engines) for full rule reasoning, coupled with the fact that symbolic inference engines only output inference results and lack unit-scale inference detection, leads to low accuracy in the generated target-sending session information. This necessitates multiple rounds of real-time communication, wasting significant communication resources, reducing communication efficiency and duration, and ultimately diminishing the user experience. Considering the following requirements for this application scenario: adaptability to unit-scale inference, high precision and accuracy requirements in the medical field, interpretable inference process output, and adaptability to various rule constraints, we have decided to adopt the following solution: In some optional implementations of certain embodiments, the above-described symbol consistency verification processing of the communication concept mapping information set to obtain a consistency verification result set, and the generation of target transmission session information and transmission to the receiving agent based on the consistency verification result set, may include the following steps: The first step, based on the communication concept mapping information set, is to perform the following conflict resolution input steps: Sub-step 1 involves performing a dimensional constraint axiom transformation on the aforementioned communication concept mapping information set and the aforementioned received ontology fragment information set to obtain a set of structural constraint mapping axioms. The structural constraint mapping axioms in this set can be derived by converting the communication concept mapping information set and the received ontology fragment information into an axiomatic structure that conforms to the input of the symbolic inferencer using an OWL axiom template. The OWL axiom template can be a template for converting equivalence relations, subordination relations, and superordinate relations into equivalence class axioms, subclass axioms, opposite-direction subclass axioms, and equivalent attribute axioms. In practice, the executing entity can first perform a mapping axiom transformation on the aforementioned communication concept mapping information set and the aforementioned received ontology fragment information set to obtain a communication mapping axiom set. Then, it can formalize the attribute and unit dimensional constraints on the communication mapping axiom set to obtain a set of structural constraint mapping axioms. The aforementioned attribute and unit dimensional constraints can be formalized by adding attribute data conversion rule information and unit dimensional constraint information to the attribute information in the communication mapping axioms.

[0063] Sub-step 2 involves modularizing the aforementioned communication task information set, communication concept mapping information set, and structural constraint mapping axiom set to obtain a locally related ontology module set. This locally related ontology module set can be a minimum subgraph set obtained by performing a breadth-first search of depth 1 to 2 using the ontology set corresponding to the communication task information set and the receiving ontology set corresponding to the communication concept mapping information set as initial nodes, and then adding the structural constraint mapping axioms corresponding to the relevant receiving ontology sets to this minimum subgraph set.

[0064] Sub-step 3 involves performing anti-pattern injection on the aforementioned local related ontology module set to obtain an anti-pattern ontology module set. The anti-pattern ontology modules in this set can be modules added to the local related ontology modules, including a set of anti-patterns containing classic logical errors. This anti-pattern set can include, but is not limited to, at least one of the following: circular subclassing pattern, mutual exclusion violation pattern, and attribute domain / scope conflict pattern.

[0065] Sub-step 4 involves transforming the aforementioned anti-pattern ontology module set into graph structure triples to obtain an ontology triple sequence. The ontology triples in this sequence can be triples where the ontology corresponding to the communication task information is the subject, the communication concept mapping information and structural edges within the module (e.g., subclass relations, equivalence relations, etc.) are the relations, and the receiving ontology is the object. This graph structure triple transformation can be achieved by first converting the ontology set included in the anti-pattern ontology module set into a graph structure and then converting it into triples.

[0066] Sub-step 5 involves inputting the aforementioned ontology triple sequence into the triple graph language model for ontology consistency verification, resulting in an ontology consistency verification information set. The ontology consistency verification information in this set can be derived from the ontology consistency verification information converted by the triple graph language model. Figure 2 The classification task yields binary classification prediction labels indicating consistency and verification confidence levels. The aforementioned triplet graph language model can be derived from the input ontology triplet sequence. Figure 2 The GLaMoR model is used for classification prediction and outputting an ontology consistency verification information set.

[0067] Sub-step 6 involves performing axiomatic consistency checks on the ontology triple sequence and structural constraint mapping axiom set corresponding to the ontology consistency check information set with inconsistent representations, resulting in an axiomatic consistency check information set. The axiomatic consistency check information in this set can be obtained by using a symbolic inference engine to perform precise consistency checks on the ontology triple sequence and structural constraint mapping axiom set corresponding to the input ontology consistency check information set with inconsistent representations using TBox (term box) and ABox (assertion box), outputting the check information for inconsistent individuals / classes and the minimal inconsistency-preserving ontology. The aforementioned inconsistent individuals / classes can refer to classes determined as "unlikely to exist as task instances" under the structural constraint mapping axiom constraints, or classes that violate logical constraints.

[0068] Sub-step 7 involves performing multi-dimensional hard constraint verification on the ontology triple sequence corresponding to the axiom consistency verification information set with inconsistent representations, thereby obtaining a hard constraint verification information set. The aforementioned multi-dimensional hard constraint verification may include, but is not limited to, at least one of the following: concept mutual exclusion verification, hierarchical inclusion verification, attribute type and value range verification, and unit dimension verification.

[0069] Sub-step 8: In response to the determination that there is an inconsistent hard constraint verification information set in the hard constraint verification information set, after conflict resolution of the inconsistent hard constraint verification information set, it is re-inputted into the semantic mapping large language model along with the corresponding target ontology set, the received ontology segment, and the current task context information to obtain the target concept mapping information set, which serves as the communication concept mapping information set, and the above conflict resolution input step is executed again. The target concept mapping information in the above target concept mapping information set can be the mapping information obtained by re-performing zero-shot mapping inference using the current task context information, the target ontology set, and the received ontology segment from the mapping information that failed verification. The conflict resolution process can involve parsing the reasons for the inconsistent verification results, the categories of conflict constraints that failed verification, conflict location information, triggering conditions, and suggested correction directions to form a structured form (e.g., JSON format).

[0070] Sub-step 9, in response to determining that the hard constraint verification information set contains a representation that has passed verification, determines the hard constraint verification information set as a consistency verification result set, and generates target sending session information and sends it to the receiving agent based on the consistency verification result set. The implementation method for generating target sending session information and sending it to the receiving agent based on the consistency verification result set can be found in the specific implementation methods given in steps 105 to 107.

[0071] The above-described technical solution and its related content, as an inventive point of this disclosure, solve the technical problem of "low accuracy of generated target-sending session information, wasting a large amount of communication resources, reducing communication efficiency and prolonging communication time, and reducing user experience." Factors leading to low accuracy of generated target-sending session information, wasted communication resources, reduced communication efficiency and prolonged communication time, and reduced user experience often include: the medical field has numerous medical compliance requirements; full rule reasoning is performed solely through symbolic inference engines (e.g., HermiT inference engine, Pellet inference engine); and symbolic inference engines only output reasoning results and lack unit-scale reasoning detection. This results in low accuracy of the generated target-sending session information, requiring multiple rounds of real-time communication, wasting a large amount of communication resources, reducing communication efficiency and prolonging communication time, and reducing user experience. Solving these factors can improve the accuracy of generated target-sending session information, reduce the waste of communication resources, improve communication efficiency and shorten communication time, and improve user experience. To achieve this effect, this disclosure firstly uses a standardized transformation of dimensional constraint axioms, mapping them to OWL axioms, adhering to the OWL2DL specification. This ensures the logical completeness of subsequent reasoning and avoids the disconnect between custom constraint rules based on unit dimensions and ontology reasoning. Secondly, ontology modularization and anti-pattern injection are employed. Ontology modularization reduces the reasoning scale by compressing the entire dataset to the smallest subgraph, effectively addressing the resource waste associated with full-scale reasoning in symbolic inference. Anti-pattern injection improves the ability to identify typical conflicts, facilitating more accurate ontology consistency checks. Next, ontology consistency prediction based on a triple graph language model improves the efficiency and accuracy of ontology checks for symbolic inference. Axiom consistency checks based on symbolic inference only re-check high-risk cases, effectively addressing inaccuracies in extreme situations. This two-level architecture balances efficiency and completeness in consistency checks, enhancing their comprehensiveness. Finally, multi-dimensional hard constraint checks systematically examine hard constraints across multiple dimensions, avoiding semantic drift caused by missing soft constraints. Then, regenerating the mapping information and re-performing symbolic consistency checks effectively solves the problem of symbolic inferencers only outputting inference results without feedback information, reducing resource waste caused by subsequent blind retries, and facilitating subsequent mapping generation and verification based on feedback information to form a recursive closed loop. Finally, generating target sending session information based on the consistency check result set and sending it to the receiving agent improves the accuracy of the consistency check result set and target sending session information, reduces communication resource waste, improves communication efficiency, shortens communication time, and enhances user experience.

[0072] Step 105: Store the communication concept mapping information set corresponding to at least one consensus verification result that has passed the characterization verification into the dynamic semantic consensus database.

[0073] In some embodiments, the aforementioned executing entity may store the communication concept mapping information set corresponding to at least one consensus verification result that has passed the representation verification into a dynamic semantic consensus database. This dynamic semantic consensus database may be a database used to store the communication concept mapping information set that has passed the representation verification at the session dimension during multi-agent communication, as well as the timeliness information of the communication concept mapping information set. This dynamic semantic consensus database can achieve version management for the reuse of communication concept mapping information.

[0074] Step 106: Based on the dynamic semantic consensus database and the communication task information set that does not meet the preset conditions, perform semantic constraint transformation on the session information to obtain the target sending session information.

[0075] In some embodiments, the aforementioned execution entity can perform semantic constraint transformation on the aforementioned session information based on the aforementioned dynamic semantic consensus database and the communication task information set that does not meet the aforementioned preset conditions, to obtain target sending session information. The target sending session information can be obtained by binding ambiguous words in the aforementioned session information to unique semantic identifiers (URIs) determined by the local ontology knowledge bases of the sending and receiving agents, and by transforming attributes by unit or scale.

[0076] As an example, the aforementioned execution entity can first, in response to determining that a set of historically verified communication task information exists in the aforementioned session information, query the aforementioned dynamic semantic consensus database to retrieve the receiving ontology set corresponding to the historical communication task information set. Then, it can annotate the communication task information sets in the aforementioned session information that do not meet the preset conditions, obtaining an annotated communication task information set. Finally, it can combine the aforementioned receiving ontology set and the annotated communication task information set to obtain the target transmission session information.

[0077] Optionally, before sending the target transmission session information to the receiving agent for reception and multi-round session communication, the method may further include the following steps: In response to the determination that there is no joint semantic drift degree satisfying the preset conditions in the aforementioned joint semantic drift degree set, the aforementioned session information is lightly annotated to obtain annotated session information, which is then sent as the target session information, or the session information is determined as the target session information. The aforementioned light annotation can be performed using a mapping table, a thesaurus, or regular expressions to semantically adjust the communication task information set corresponding to joint semantic drift degrees that do not satisfy the preset conditions (e.g., non-standard aliases, different units, lack of explicit type declarations), thereby obtaining annotated session information, which is then sent as the target session information.

[0078] Step 107: Send the target sending session information to the receiving agent so that the receiving agent can receive it and conduct multi-round session communication.

[0079] In some embodiments, the aforementioned executing entity may send the aforementioned target sending session information to the aforementioned receiving agent, so that the receiving agent can receive it and conduct multi-round session communication. Therefore, this application does not force the sending agent and receiving agent in communication to adopt the same unified global model, but allows each agent to retain its own ontology, terminology system (i.e., local ontology knowledge base), and task perspective. During communication, the concepts included in the session information can be locally negotiated and dynamically aligned as needed, so that even when new concepts, new attributes, or cross-domain tasks appear during communication, the consistency of long-term collaboration can still be maintained through local semantic reconstruction.

[0080] The above embodiments of this disclosure have the following beneficial effects: The multi-agent communication method based on ontology semantic alignment in some embodiments of this disclosure is suitable for real-time, multi-round communication between multiple agents, improving the accuracy and efficiency of multi-agent communication collaboration, reducing the waste of communication transmission resources, and shortening the communication collaboration time. Specifically, the reasons why it is difficult to adapt to multi-round real-time scenarios, increasing the consumption of communication transmission resources, reducing the accuracy and efficiency of multi-agent communication, and prolonging the communication collaboration time are as follows: Multi-agent communication using fixed communication structure protocols or static ontology is difficult to handle temporarily emerging new concepts and attributes in multi-round real-time communication within a multi-agent system, and structural mismatch problems are prone to occur in cross-domain multi-agent communication collaboration; simultaneously, all communication information between multiple agents requires high-cost negotiation and repeated multi-round communication negotiations regarding concept conflicts, leading to increased consumption of communication transmission resources, reduced accuracy and efficiency of multi-agent communication, and prolonged communication collaboration time. Based on this, some embodiments of the multi-agent communication method based on ontology semantic alignment disclosed herein can first extract communication task information from the session information to be sent by the sending agent in the multi-agent communication system to obtain a communication task information set. The multi-agent communication system further includes a receiving agent and a dynamic semantic consensus database. Here, by extracting information before sending the session information, semantic drift suppression can be brought forward to the entry point of the execution chain, avoiding further expansion of erroneous semantic information. Secondly, a joint semantic drift degree set is determined from the aforementioned communication task information set, the receiving agent's local ontology knowledge base, and the current task context information. Determining the joint semantic drift degree set facilitates the subsequent execution of different processing methods, avoiding high-cost negotiation communication for all session information, balancing communication accuracy and real-time performance, and reducing the overhead of communication and computing resources. Thirdly, in response to determining that at least one joint semantic drift degree satisfying a preset condition exists in the aforementioned joint semantic drift degree set, a communication concept mapping information set is generated using a semantic mapping large language model based on the communication task information set corresponding to the at least one joint semantic drift degree, the receiving local ontology knowledge base, and the current task context information. Here, a communication concept mapping information set is generated from the communication task information set that meets preset conditions. A hierarchical control strategy of lightweight detection and on-demand semantic alignment is employed, along with the use of a semantic mapping large language model to improve the accuracy of the communication concept mapping information set. Next, the communication concept mapping information set undergoes symbolic consistency verification to obtain a consistency verification result set. This symbolic consistency verification process reduces the illusionary alignment generated by the semantic mapping large language model, combining the flexibility of a neural model with the reliability of formal logic. Subsequently, the communication concept mapping information set corresponding to at least one consistency verification result that passes the representation verification is stored in a dynamic semantic consensus database.Here, the dynamic semantic consensus database can be used to store verified communication concept mapping information, enabling the reuse of the same semantics in multi-round communication, reducing redundant negotiation communication between multiple agents, reducing communication resource consumption and the time required for subsequent semantic transformation, and improving the efficiency of multi-agent communication. Then, based on the dynamic semantic consensus database and the communication task information set that does not meet the above preset conditions, the above session information is semantically constrained and transformed to obtain the target sending session information. Here, the efficiency of semantic constraint transformation and the quality of the target sending session information can be improved, flexibly adapting to the emergence of new concepts and attributes during communication, and effectively reducing the probability of semantic offset problems. Finally, the above target sending session information is sent to the above receiving agent for reception and multi-round session communication. Here, the waste of communication transmission resources can be reduced, communication efficiency and real-time performance can be improved, and the duration of multi-agent communication collaboration can be reduced. Therefore, the multi-agent communication method based on ontology semantic alignment is suitable for real-time multi-round communication among multiple agents. It can promptly detect semantic deviations in the communication process, trigger semantic alignment processing of communication concept mapping and consistency verification of alignment results as needed, and enable agents to cache and reuse alignment results in multi-round communication while maintaining the independence of their respective ontology knowledge bases. This can improve the accuracy and efficiency of multi-agent communication collaboration, reduce the waste of communication transmission resources, and shorten the communication collaboration time.

[0081] Further reference Figure 2 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of a multi-agent communication device based on ontology semantic alignment. These device embodiments are similar to... Figure 1 Corresponding to the method embodiments shown, this ontology-based semantic alignment multi-agent communication device can be specifically applied to various electronic devices.

[0082] like Figure 2As shown, a multi-agent communication device 200 based on ontology semantic alignment includes: an information extraction unit 201, a determination unit 202, a generation unit 203, a symbol consistency verification unit 204, a storage unit 205, a semantic constraint conversion unit 206, and a sending unit 207. The information extraction unit 201 is configured to extract communication task information from the session information to be sent by the sending agent in the multi-agent communication system to obtain a communication task information set. The multi-agent communication system further includes a receiving agent and a dynamic semantic consensus database. The determination unit 202 is configured to determine the joint semantic drift set of the communication task information set, the receiving agent's local ontology knowledge base, and the current task context information. The generation unit 203 is configured to, in response to determining that at least one joint semantic drift degree satisfying a preset condition exists in the joint semantic drift set, generate a communication concept mapping information set using a semantic mapping large language model, based on the communication task information set corresponding to the at least one joint semantic drift degree, the receiving local ontology knowledge base, and the current task context information. Symbol consistency verification unit 204 is configured to perform symbol consistency verification on the aforementioned communication concept mapping information set to obtain a consistency verification result set. Storage unit 205 is configured to store the communication concept mapping information set corresponding to at least one consistency verification result that passed the verification into the dynamic semantic consensus database. Semantic constraint transformation unit 206 is configured to perform semantic constraint transformation on the aforementioned session information based on the aforementioned dynamic semantic consensus database and the communication task information set that does not meet the aforementioned preset conditions to obtain target transmission session information. Transmission unit 207 is configured to transmit the aforementioned target transmission session information to the aforementioned receiving agent for reception and multi-round session communication.

[0083] It is understandable that the units and references described in the multi-agent communication equipment 200 based on ontology semantic alignment Figure 1 The steps in the described method correspond to each other. Therefore, the operations, features, and beneficial effects described above for the method are also applicable to the ontology-based semantic alignment multi-agent communication equipment 200 and the units contained therein, and will not be repeated here.

[0084] The following is for reference. Figure 3 It shows a schematic diagram of the structure of an electronic device (e.g., an electronic device) 300 suitable for implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0085] like Figure 3As shown, the electronic device 300 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the electronic device 300. The processing device 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0086] Typically, the following devices can be connected to I / O interface 305: input devices 306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows electronic device 300 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 3 An electronic device 300 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 3 Each box shown can represent a device or multiple devices as needed.

[0087] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 309, or installed from storage device 308, or installed from ROM 302. When the computer program is executed by processing device 301, it performs the functions defined above in the methods of some embodiments of this disclosure.

[0088] It should be noted that, in some embodiments of this disclosure, the computer-readable storage medium described above can be, for example,—but is not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, the computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0089] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0090] The aforementioned computer-readable storage medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to: extract communication task information from the session information to be sent by the sending agent in the multi-agent communication system to obtain a communication task information set, wherein the aforementioned multi-agent communication system further includes: a receiving agent and a dynamic semantic consensus database; determine a joint semantic drift set of the aforementioned communication task information set, the receiving agent's local ontology knowledge base, and the current task context information; in response to determining that at least one joint semantic drift satisfying a preset condition exists in the aforementioned joint semantic drift set, utilize a semantic mapping large language model, based on the aforementioned at least one... A communication concept mapping information set is generated by combining the communication task information set corresponding to the joint semantic drift degree, the aforementioned local ontology knowledge base, and the aforementioned current task context information; the aforementioned communication concept mapping information set is subjected to symbol consistency verification processing to obtain a consistency verification result set; the communication concept mapping information set corresponding to at least one consistency verification result that passes the representation verification is stored in the dynamic semantic consensus database; based on the aforementioned dynamic semantic consensus database and the communication task information set that does not meet the aforementioned preset conditions, the aforementioned session information is subjected to semantic constraint transformation to obtain target sending session information; the aforementioned target sending session information is sent to the aforementioned receiving agent for the receiving agent to receive and conduct multi-round session communication.

[0091] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0092] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0093] The units described in some embodiments of this disclosure can be implemented in software or hardware. The described units can also be housed in a processor; for example, a processor may be described as including an information extraction unit, a determination unit, a generation unit, a symbol consistency verification unit, a storage unit, a semantic constraint conversion unit, and a transmission unit. The names of these units do not necessarily limit the specific unit itself; for example, the extraction unit may also be described as "a unit that extracts communication task information from the session information to be sent by the sending agent in a multi-agent communication system to obtain a communication task information set."

[0094] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0095] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A multi-agent communication method based on ontology semantic alignment, comprising: Communication task information is extracted from the session information to be sent by the sending agent in the multi-agent communication system to obtain a communication task information set. The multi-agent communication system further includes a receiving agent and a dynamic semantic consensus database. Determine the joint semantic drift set of the communication task information set, the receiving agent's local ontology knowledge base, and the current task context information; In response to determining that at least one joint semantic drift degree in the joint semantic drift degree set satisfies a preset condition, a communication concept mapping information set is generated using a semantic mapping large language model based on the communication task information set corresponding to the at least one joint semantic drift degree, the receiving local ontology knowledge base, and the current task context information; The symbol consistency verification process is performed on the communication concept mapping information set to obtain a consistency verification result set; Store the communication concept mapping information set corresponding to at least one consensus verification result that has passed the characterization verification into the dynamic semantic consensus database; Based on the dynamic semantic consensus database and the communication task information set that does not meet the preset conditions, the session information is semantically constrained to obtain the target sending session information; The target sending session information is sent to the receiving agent so that the receiving agent can receive it and conduct multi-round session communication.

2. The method according to claim 1, wherein, Before sending the target session information to the receiving agent for reception and multi-round session communication, the method further includes: In response to determining that there is no joint semantic drift degree that satisfies a preset condition in the set of joint semantic drift degrees, the session information is lightly labeled to obtain labeled session information, which is then sent as the target session information, or the session information is determined as the target session information.

3. The method according to claim 1, wherein, The step of extracting communication task information from the session information to be sent by the sending agent in the multi-agent communication system to obtain a communication task information set includes: The conversation information is preprocessed to obtain preprocessed conversation information; The preprocessed session information is segmented to obtain a communication segmentation set; Dependency parsing is performed on the communication word segmentation set to obtain a conversation dependency parsing tree; The initial concept information set is obtained by performing rule matching and extraction on the session dependency syntax tree. The initial concept information set is subjected to coreference resolution to obtain the communication task information set.

4. The method according to claim 1, wherein, The determination of the joint semantic drift set of the communication task information set, the receiving agent's local ontology knowledge base, and the current task context information includes: Based on the message header information in the session information, determine the receiving agent's local ontology knowledge base; Perform multi-granularity matching on the communication task information set and the local ontology knowledge base to obtain a first matching receiver ontology set; The communication task information set and the local ontology knowledge base are matched by proximity embedding to obtain a second matched receiving ontology set. Determine the multi-dimensional semantic matching degree set of the first matching receiver ontology set, the second matching receiver ontology set, and the communication task information set; Determine the multidimensional context consistency value set of the communication task information set and the current task context information; The set of differences between the preset value and the weighted sum of each multidimensional semantic matching degree in the multidimensional semantic matching degree set and the corresponding multidimensional context consistency degree value in the multidimensional context consistency degree value set is determined as the joint semantic drift degree set.

5. The method according to claim 1, wherein, The step of utilizing a semantic mapping large language model to generate a communication concept mapping information set based on the communication task information set corresponding to at least one joint semantic drift degree, the received local ontology knowledge base, and the current task context information includes: The communication task information set corresponding to the at least one joint semantic drift degree is determined as the target ontology set in the local ontology knowledge base corresponding to the sending agent, wherein the target ontology set includes: a concept set, a relation set, and a logical constraint information set; The target ontology set, the corresponding receiving ontology segment in the receiving local ontology knowledge base, and the current task context information are processed in a structured manner to obtain structured input text information; Based on the structured input text information, generate zero-sample prompt word information; The zero-sample prompt word information is input into the semantic mapping large language model to perform zero-sample inference, thereby obtaining a candidate concept mapping information set, wherein the candidate concept mapping information set includes at least one of the following: mapping type information, mapping confidence, attribute correspondence information, and attribute transformation function; The candidate concept mapping information set is subjected to conflict classification processing to obtain the communication concept mapping information set.

6. The method according to claim 1, wherein, The symbol consistency verification process performed on the communication concept mapping information set to obtain a consistency verification result set includes: Based on the communication concept mapping information set, perform the following input steps: The communication concept mapping information set and the logical constraint information set included in the received local ontology knowledge base are converted into a reasoning format to obtain a formula reasoning triple set. A multidimensional logical consistency check is performed on the formula reasoning triple set to obtain a multidimensional logical consistency check information set. The multidimensional logical consistency check includes at least one of the following: checks to determine whether the candidate concept mapping information triggers a concept unsatisfiability check, a hierarchical relationship conflict check, an attribute domain and value domain mismatch check, a numerical out-of-bounds check, or a unit dimension incommutation check. In response to the determination that successful verification information exists in the multidimensional logical consistency verification information set, the successful multidimensional logical consistency verification information is determined as the consistency verification result set; In response to the determination that there is verification information with failed representation in the multidimensional logical consistency verification information set, constraint reasoning parsing is performed on the multidimensional logical consistency verification information set with failed representation to obtain the verification parsing information set; Based on the verification and parsing information set, a new communication remapping information set is generated as a communication concept mapping information set, so as to execute the input step again.

7. The method according to claim 6, wherein, The step of regenerating the communication remapping information set based on the verification and parsing information set, as the communication concept mapping information set, to perform the input step again, includes: Based on each multi-dimensional logical consistency verification information in the multi-dimensional logical consistency verification information set corresponding to the verification parsing information set, the following verification steps are performed: Determine the number of failed checks performed on the multi-dimensional logical consistency verification information; Perform symbolic consistency verification on the verification parsing information corresponding to the multidimensional logical consistency verification information to obtain the target verification information; In response to the determination that the target verification information represents successful verification, the communication concept mapping information corresponding to the target verification information is stored in the dynamic semantic consensus database; In response to the determination that the target verification information indicates a verification failure, and the number of failed verification executions exceeds a preset execution threshold, the execution of symbol consistency verification is terminated; In response to the determination that the target verification information indicates verification failure, and the number of failed verification executions is less than or equal to the preset execution threshold, the verification parsing information, the corresponding target ontology set, the received ontology segment, and the current task context information are input into the semantic mapping large language model to re-map concepts and obtain communication remapping information, which serves as the communication concept mapping information set, so as to execute the input step again.

8. A multi-agent communication device based on ontology semantic alignment, comprising: The information extraction unit is configured to extract communication task information from the session information to be sent by the sending agent in the multi-agent communication system to obtain a communication task information set. The multi-agent communication system further includes a receiving agent and a dynamic semantic consensus database. The determining unit is configured to determine the joint semantic drift set of the communication task information set, the receiving agent's local ontology knowledge base, and the current task context information; The generation unit is configured to, in response to determining that there exists at least one joint semantic drift degree satisfying a preset condition in the joint semantic drift degree set, utilize a semantic mapping large language model to generate a communication concept mapping information set based on the communication task information set corresponding to the at least one joint semantic drift degree, the received local ontology knowledge base, and the current task context information; The symbol consistency verification unit is configured to perform symbol consistency verification processing on the communication concept mapping information set to obtain a consistency verification result set. The storage unit is configured to store the communication concept mapping information set corresponding to at least one consensus verification result that has passed the characterization verification into the dynamic semantic consensus database; The semantic constraint conversion unit is configured to perform semantic constraint conversion on the session information based on the dynamic semantic consensus database and the communication task information set that does not meet the preset conditions, so as to obtain the target sending session information; The sending unit is configured to send the target sending session information to the receiving agent, so that the receiving agent can receive the information and conduct multi-round session communication.

9. An electronic device, comprising: One or more processors; Storage device, on which one or more programs are stored, When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, wherein, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-7.

Images