Text processing method and related device

Abstract

This disclosure provides a text processing method and related equipment, relating to the field of text processing technology. The method includes: assigning multiple segments of the text to be processed, obtained by dividing it into segments, to representative agents of various dynamic consensus groups for processing, obtaining processing results from each representative agent; determining the confidence score of each representative agent's processing result; determining the fusion priority of each representative agent's processing result based on the confidence score; and merging the processing results of each representative agent based on the fusion priority to obtain the parsed result of the text to be processed. Through the above technical means, the problems of centralized bottlenecks, outdated information fusion mechanisms, and poor scalability in multi-agent text processing in related technologies are solved.

Description

Technical Field

[0001] This disclosure relates to the field of text processing technology, and in particular to a text processing method and related equipment. Background Technology

[0002] With the improvement of large language models, intelligent agents have shown great potential in processing extremely long texts. However, single agents are limited by context length, inference costs, and the risk of illusion, making it difficult to guarantee the completeness and accuracy of the output. Related technologies improve efficiency and robustness through task decomposition and parallel processing of multiple agents.

[0003] Related technologies often employ a centralized architecture, introducing a coordinator or leader for task scheduling and result aggregation. However, these technologies suffer from the following significant drawbacks: reliance on a single coordinator leads to centralized bottlenecks, easily resulting in single points of failure and communication congestion; information fusion mechanisms are outdated, often employing simple splicing or priority overlay, making it difficult to identify illusory or outdated information; and scalability is poor, requiring new agents to be re-registered with the central node, hindering dynamic self-organization and elastic expansion. Summary of the Invention

[0004] This disclosure provides a text processing method and related apparatus to improve the reliability, consistency and scalability of text processing.

[0005] According to one aspect of this disclosure, a text processing method is provided, comprising: assigning multiple segments obtained by dividing the text to be processed to representative agents of each dynamic consensus group for processing, thereby obtaining processing results of each representative agent; determining a confidence score for the processing results of each representative agent; determining a fusion priority for the processing results of each representative agent based on the confidence scores; and merging the processing results of each representative agent based on the fusion priority to obtain a parsed result of the text to be processed.

[0006] In one embodiment of this disclosure, before assigning multiple segments of the text to be processed to representative agents of each dynamic consensus group for processing and obtaining the processing results of each representative agent, the method further includes: extracting semantic features of the task description of each agent; determining the semantic distance between the semantic features corresponding to any two agents; based on the semantic distance between the semantic features corresponding to any two agents, using a clustering algorithm to divide the multiple agents into multiple dynamic consensus groups, and determining the representative agent of the dynamic consensus group from one or more agents contained in each dynamic consensus group.

[0007] In one embodiment of this disclosure, multiple segments obtained by dividing the text to be processed are assigned to representative agents of each dynamic consensus group for processing, and the processing results of each representative agent are obtained. This includes: querying the context-aware memory pool by each representative agent to obtain the memory entries on which the segments processed by each representative agent depend; processing the segments by each representative agent based on the queried memory entries to obtain the processing results of each representative agent, and generating the dependency relationship between the segments processed by each representative agent and the memory entries on which the segments depend.

[0008] In one embodiment of this disclosure, determining the confidence score of the processing results of each representative agent includes: acquiring user input information and determining the semantic matching degree between the processing results of each representative agent and the user input information; determining the source type of the segment processed by each agent and determining the data source strength of the processing results of each representative agent based on the source type of the segment processed by each agent; and determining the confidence score of the processing results of each representative agent based on the confidence score, data source strength, and semantic matching degree corresponding to the processing results of each representative agent.

[0009] In one embodiment of this disclosure, the fusion priority of the processing results of each representative agent is determined based on the confidence score of the processing results of each representative agent, including: obtaining the generation timestamp of the processing results of each representative agent; and determining the fusion priority of the processing results of each representative agent based on the confidence score and generation timestamp of the processing results of each representative agent.

[0010] In one embodiment of this disclosure, the processing results of each representative agent are merged based on the fusion priority of their processing results to obtain the parsing result of the text to be processed. This includes: performing conflict detection and / or consistency verification on the processing results of any two representative agents; when the processing results of any two representative agents pass the conflict detection and / or consistency verification, the processing results of each representative agent are merged based on the fusion priority of their processing results to obtain the parsing result.

[0011] In one embodiment of this disclosure, conflict detection is performed on the processing results of any two representative agents, including: detecting whether the difference in confidence scores of the processing results of any two representative agents is greater than a preset threshold; if it is greater, then it is determined that the processing results of the two representative agents fail the conflict detection; or detecting whether the difference in generation timestamps of the processing results of any two representative agents is less than the larger of the effective lifetimes of the processing results of the two representative agents; if it is less, then it is determined that the processing results of the two representative agents fail the conflict detection.

[0012] In one embodiment of this disclosure, detecting whether the difference between the generation timestamps of the processing results of any two representative agents is less than the larger of the effective lifetimes of the processing results of the two representative agents includes: determining the effective lifetime of the processing results of each representative agent based on the confidence score and content length of the processing results of each representative agent.

[0013] In one embodiment of this disclosure, the effective lifetime of the processing results of each representative agent is determined based on the confidence score and content length of the processing results of each representative agent, including: determining the base validity period, the confidence decay adjustment factor, and the content length gain coefficient; and determining the effective lifetime of the processing results of each representative agent based on the base validity period, the confidence decay adjustment factor, the content length gain coefficient, the confidence score, and the content length of the processing results of each representative agent.

[0014] In one embodiment of this disclosure, the consistency check of the processing results of any two representative agents includes: detecting whether there is a semantic conflict in the processing results of any two representative agents; if so, determining that the processing results of the two representative agents do not pass the consistency check.

[0015] In one embodiment of this disclosure, the method further includes: if the processing results of any two representative agents fail the consistency check, then backtracking the memory entries on which the segments processed by the two representative agents depend, in order to locate and correct the source of the error.

[0016] In one embodiment of this disclosure, multiple segments obtained by dividing the text to be processed are assigned to representative agents of each dynamic consensus group for processing to obtain the processing results of each representative agent. This includes: querying the context-aware memory pool by each representative agent to obtain the historical processing information of each representative agent, wherein the historical processing information of each agent includes multiple memory entries; processing the segments by each representative agent based on the historical processing information of each representative agent to obtain the processing results of each representative agent, and generating the dependency relationship between the segments processed by each representative agent and the memory entries on which the segments depend.

[0017] In one embodiment of this disclosure, each memory entry includes at least one of the following: the agent's identifier, the identifier of the memory entry, the processed text or fragment, a list of identifiers of memory entries on which the processed text or fragment depends, the processing result, and the confidence score of the processing result, the survival time, and the generation timestamp.

[0018] In one embodiment of this disclosure, the context-aware memory pool provides a multi-dimensional query interface, which includes at least one of the following dimensions: keyword-based query, time window-based query, and confidence score-based query.

[0019] In one embodiment of this disclosure, memory entries with a lifetime greater than the effective lifetime are marked as invalid in the context-aware memory pool; memory entries with a confidence score less than a preset score are also marked as invalid in the context-aware memory pool.

[0020] According to another aspect of this disclosure, a text processing apparatus is provided, comprising: a processing unit configured to assign multiple segments obtained by dividing a text to be processed to representative agents of each dynamic consensus group for processing, thereby obtaining processing results from each representative agent; a first determining unit configured to determine a confidence score of the processing results from each representative agent; a second determining unit configured to determine a fusion priority of the processing results from each representative agent based on the confidence scores of the processing results from each representative agent; and a merging unit configured to merge the processing results from each representative agent based on the fusion priority of the processing results from each representative agent, thereby obtaining a parsed result of the text to be processed.

[0021] According to another aspect of this disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform any of the methods described above by executing the executable instructions.

[0022] According to another aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements any of the methods described above.

[0023] According to another aspect of this disclosure, a computer program product is provided, including computer instructions stored in a computer-readable storage medium, which, when executed by a processor, implement operation instructions for any of the methods described above.

[0024] In the embodiments of this disclosure, multiple segments obtained by dividing the text to be processed are assigned to representative agents of each dynamic consensus group for processing. The confidence score of the processing results of each representative agent is determined, and then the fusion priority of the processing results of each representative agent is determined. Finally, the processing results of each representative agent are fused together. This solves the problems of centralized bottleneck, backward information fusion mechanism and poor scalability in multi-agent text processing in related technologies, so as to improve the reliability, consistency and scalability of text processing.

[0025] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0026] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0027] Figure 1 A schematic diagram of a text processing system according to an embodiment of the present disclosure is shown.

[0028] Figure 2 A flowchart of a text processing method according to an embodiment of this disclosure is shown.

[0029] Figure 3 A flowchart of a dynamic consensus group partitioning method is shown in an embodiment of this disclosure.

[0030] Figure 4 A flowchart illustrating a confidence score determination method according to an embodiment of this disclosure is shown.

[0031] Figure 5 A flowchart of a fusion priority determination method is shown in an embodiment of this disclosure.

[0032] Figure 6 A flowchart illustrating a method for determining effective survival time in an embodiment of this disclosure is shown.

[0033] Figure 7 A flowchart of another dynamic consensus group partitioning method in an embodiment of this disclosure is shown.

[0034] Figure 8 A schematic diagram illustrating memory entry details in an embodiment of this disclosure is shown.

[0035] Figure 9 A flowchart illustrating another text processing method in an embodiment of this disclosure is shown.

[0036] Figure 10 A schematic diagram of a text processing apparatus according to an embodiment of the present disclosure is shown.

[0037] Figure 11 A schematic diagram of an electronic device provided in an embodiment of this disclosure is shown. Detailed Implementation

[0038] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that this disclosure will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0039] Furthermore, the accompanying drawings are merely illustrative of this disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0040] It should be understood that the steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this disclosure is not limited in this respect.

[0041] 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.

[0042] 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".

[0043] It should be noted that, unless otherwise specified, the embodiments of this disclosure and the technical features thereof can be combined with each other.

[0044] To facilitate understanding, the following is an explanation of several terms used in this disclosure:

[0045] Greedy Clustering Algorithm: A clustering method based on a local optimum selection strategy. It iteratively merges the most similar agents or data points into the same group to quickly form a cluster structure. It is often used to build dynamic consensus groups and has the characteristics of high computational efficiency and suitability for real-time scenarios.

[0046] Byzantine Fault Tolerant Protocol (BFT): A type of distributed consensus protocol that can ensure that honest nodes in the system reach consensus even when malicious or faulty nodes send erroneous information. It is suitable for multi-agent collaborative environments with high security requirements.

[0047] Raft-like Protocol: A type of protocol that draws on the design ideas of the Raft consensus algorithm, emphasizing leader election, log replication and security mechanisms. It achieves strong consistency through a clear master-slave role division and is suitable for intra-group coordination and state synchronization.

[0048] Raft (Replication and Fault Tolerance): A consistency algorithm for managing replicated logs in distributed systems. It achieves strong consistency through explicit leader election, log replication, and security mechanisms. Its design emphasizes understandability and implementability, and it is suitable for intra-group state synchronization and coordination control in multi-agent systems.

[0049] BFT voting: A voting process based on the Byzantine fault tolerance mechanism. Each node submits its voting opinion in the consensus decision-making process. The system reaches a reliable consensus by verifying the legality of the vote and counting the majority results. Even if some participants behave abnormally, the correctness of the decision can still be guaranteed.

[0050] Hierarchical Clustering: An algorithm that constructs a tree-like clustering structure by recursively merging close clusters (agglomerative) or splitting large clusters (split-like). It can discover multi-level organizational relationships in data and is suitable for dynamic consensus group partitioning scenarios where there is no need to pre-determine the number of clusters.

[0051] K-means (K-means Clustering): A distance-based iterative clustering algorithm that divides data into K clusters by minimizing the sum of the squared distances from each point to the center of its cluster. It is computationally efficient and suitable for large-scale data, and is often used for grouping semantic vectors in intelligent agent tasks.

[0052] DBSCAN (Density-Based Spatial Clustering of Applications with Noise): A density-based clustering algorithm that discovers clusters of arbitrary shapes by identifying high-density regions and separating low-density regions. It can also identify noise points and is suitable for processing semantic feature data of agents that are unevenly distributed or contain outliers.

[0053] A token is a basic semantic unit for segmenting text in natural language processing. A token can include words, subwords, or character-level fragments.

[0054] Figure 1 This illustration shows a schematic diagram of a text processing system according to an embodiment of the present disclosure. The text processing system includes:

[0055] Perception layer 101, memory layer 102, consensus layer 103 and fusion layer 104.

[0056] The perception layer 101 contains multiple agents: agent A, agent B, agent C, and agent D.

[0057] Memory layer 102 is a shared context-aware memory pool. This pool stores and manages memory entries generated by each agent, supporting dependency tracking, temporal evolution, and semantic association. It enables unified management of historical information and context-aware capabilities.

[0058] Consensus layer 103 includes dynamic consensus group 1 and dynamic consensus group 2. These two dynamic consensus groups can be formed based on information from the memory layer. The two consensus groups communicate and coordinate through a BFT voting channel to achieve highly reliable consensus decisions. The BFT mechanism ensures that consensus can still be reached even in the presence of anomalies or malicious behavior.

[0059] The fusion layer 104 contains a fusion engine. This fusion engine comprehensively considers the confidence score and generation timestamp of each processing result and performs weighted fusion of results from different consensus groups.

[0060] The text processing system may include an application program that performs the following actions: assigning multiple segments of the text to be processed to representative agents of each dynamic consensus group for processing, obtaining the processing results of each representative agent; determining the confidence score of the processing results of each representative agent; determining the fusion priority of the processing results of each representative agent based on the confidence score of the processing results of each representative agent; and merging the processing results of each representative agent based on the fusion priority of the processing results of each representative agent to obtain the parsed result of the text to be processed.

[0061] Figure 2 This diagram illustrates a flowchart of a text processing method according to an embodiment of the present disclosure, the method as follows: Figure 2 As shown, it includes the following steps:

[0062] S201, the multiple segments obtained from dividing the text to be processed are assigned to the representative agents of each dynamic consensus group for processing, and the processing results of each representative agent are obtained.

[0063] As an example, the text to be processed is the original input document that needs to be processed.

[0064] As an example, each dynamic consensus group is obtained by dividing agents using a clustering algorithm. Each dynamic consensus group includes one or more agents, and the agents in each dynamic consensus group have similar analytical capabilities.

[0065] As an example, one agent in each dynamic consensus group is the representative agent of that dynamic consensus group. The representative agent of each dynamic consensus group is the center or leader of that dynamic consensus group, determined by a clustering algorithm, and is the most representative agent.

[0066] As an example, the processing results represent the output generated by the agent after performing semantic analysis on the assigned fragments, including extracted summaries and analyses.

[0067] S202, determine the confidence score of the processing results of each representative agent.

[0068] As an example, the confidence score is used to measure the reliability of the processing results.

[0069] S203, based on the confidence scores of the processing results of each representative agent, determine the fusion priority of the processing results of each representative agent.

[0070] As an example, fusion prioritization is used to measure the importance of processing results and the order in which they are merged.

[0071] S204. Based on the fusion priority of the processing results of each representative agent, the processing results of each representative agent are merged to obtain the parsing result of the text to be processed.

[0072] In this embodiment, the text to be processed is divided into multiple segments and assigned to representative agents of each dynamic consensus group. After each representative agent generates a processing result, a confidence score is determined for each processing result, and a fusion priority is determined based on the confidence score. Then, all processing results are merged according to the fusion priority to finally generate the parsed result of the text to be processed. Even if one or some representative agents fail, other representative agents can still be used, avoiding the single point of failure risk brought by a centralized coordinator; fusion results are merged according to the fusion priority to improve the fusion effect; if a new agent is added, the representative agents of the original dynamic consensus group are updated, or a new dynamic consensus group adds a new representative agent. Therefore, through the above technical means, the problems of centralized bottleneck, outdated information fusion mechanism and poor scalability in multi-agent text processing in related technologies are solved, thereby improving the reliability, consistency and scalability of text processing.

[0073] Figure 3 This invention discloses a flowchart of a dynamic consensus group partitioning method according to an embodiment of the present disclosure. The method is as follows: Figure 3As shown, it includes the following steps:

[0074] S301, Extract the semantic features of the task descriptions of each agent;

[0075] S302, determine the semantic distance between the semantic features corresponding to any two agents;

[0076] S303, based on the semantic distance between the semantic features corresponding to any two agents, uses a clustering algorithm to divide multiple agents into multiple dynamic consensus groups, and determines the representative agent of each dynamic consensus group from one or more agents contained in each dynamic consensus group.

[0077] An exemplary task description is descriptive information about the tasks that an agent can handle.

[0078] Exemplary semantic features are vectorized representations of task descriptions.

[0079] As an example, semantic distance is used to measure the degree of difference between two semantic features, and can be cosine distance or Euclidean distance.

[0080] Exemplary clustering algorithms are computational methods for grouping multiple agents based on semantic distance. These algorithms include hierarchical clustering, K-means, DBSCAN, or greedy clustering algorithms.

[0081] In this embodiment, the semantic distance between the semantic features extracted by any two agents is calculated.

[0082] In this embodiment, the semantic distance between the semantic features corresponding to any two agents is determined. Based on the semantic distance, a clustering algorithm is used to divide multiple agents into multiple dynamic consensus groups, and a representative agent of each dynamic consensus group is determined (the representative agent is the agent in the dynamic consensus group whose processing result is closest to the cluster center). Through the above technical means, multiple representative agents are determined, thereby achieving decentralization.

[0083] In one exemplary embodiment, the semantic features of the second text as follows:

[0084] (1);

[0085] in, For the BERT model, The second text is the task description or processing text for the i-th agent. For the embedded dimension.

[0086] The semantic distance between any two semantic features extracted by the agents can be calculated using the following formula:

[0087] (2);

[0088] Set threshold ,like If so, then agents i and j are considered to belong to the same semantic domain.

[0089] The agents are divided into several groups using a greedy clustering algorithm. Each group satisfies:

[0090] (3);

[0091] Each group elects a representative intelligent agent (Leader), and the state synchronization within the group is achieved using the Raft-like protocol.

[0092] Inter-group communication uses a Byzantine Fault Tolerance (BFT) protocol to ensure global consistency even when some nodes fail.

[0093] In an optional embodiment, the method further includes, before obtaining the processing results of each representative agent from the dynamic consensus group, assigning multiple segments of the text to be processed to representative agents of each agent for processing; acquiring historical processing records of multiple agents, including processing results generated by each agent in past tasks and corresponding user feedback; calculating feedback scores of each agent based on the historical processing records of each agent, including a weighted composite value of task completion accuracy, response latency, and error correction frequency; and determining the agent with the highest feedback score in each dynamic consensus group as the representative agent.

[0094] In one embodiment of this disclosure, multiple segments obtained by dividing the text to be processed are assigned to representative agents of each dynamic consensus group for processing, and the processing results of each representative agent are obtained. This includes: querying the context-aware memory pool by each representative agent to obtain the memory entries on which the segments processed by each representative agent depend; processing the segments by each representative agent based on the queried memory entries to obtain the processing results of each representative agent, and generating the dependency relationship between the segments processed by each representative agent and the memory entries on which the segments depend.

[0095] As an example, a context-aware memory pool is a shared storage structure used to store and manage the historical processing information of each agent, which includes multiple memory entries.

[0096] Exemplary dependency relationships include the following types: semantic continuation, factual reference, logical reasoning, entity association, chronological sequence, and modification / update. Semantic continuation indicates that the meaning of a fragment depends on a concept or proposition in a memory entry; factual reference indicates that the fragment directly references specific facts or data from a memory entry; logical reasoning indicates that the content of the fragment is a conclusion drawn from reasoning based on a memory entry; entity association indicates that the entities involved in the fragment have a referential, inclusive, or parallel relationship with entities in the memory entry; chronological sequence indicates that the event described in the fragment is later in time than the event in the memory entry it depends on; and modification / update indicates that the fragment modifies or updates the information in the memory entry.

[0097] In this embodiment, each representative agent queries a context-aware memory pool for memory entries related to the content of the segment it is processing, serving as context information for that segment. Subsequently, the representative agent performs semantic analysis and reasoning on the current text segment based on the retrieved memory entries, generating a processing result. During the processing, the agent records the dependencies between the text segment and the memory entries it relies on, thus ensuring the traceability of the processing. This technical approach addresses the problems of insufficient context sharing among agents and a lack of effective access to intermediate information in related technologies.

[0098] In one optional embodiment, the process includes: each representative agent retrieving a set of memory entries from a context-aware memory pool based on the feature vector of its assigned segment; using the retrieved memory entries as context information, combining them with the current segment for semantic expansion and logical reasoning to generate the processing result for that segment; and simultaneously recording the mapping relationship between the segment and the used memory entries, forming a dependency relationship between the segment and its dependent memory entries. Through the above technical means, context sharing among multiple agents is achieved.

[0099] In one embodiment of this disclosure, multiple segments obtained by dividing the text to be processed are assigned to representative agents of each dynamic consensus group for processing to obtain the processing results of each representative agent. This includes: querying the context-aware memory pool by each representative agent to obtain the historical processing information of each representative agent, wherein the historical processing information of each agent includes multiple memory entries; processing the segments by each representative agent based on the historical processing information of each representative agent to obtain the processing results of each representative agent, and generating the dependency relationship between the segments processed by each representative agent and the memory entries on which the segments depend.

[0100] In this embodiment, each representative agent queries the context-aware memory pool to obtain historical processing information from itself and other agents. Using the retrieved memory entries as context information, semantic parsing and reasoning are performed on the assigned text fragments to generate corresponding processing results. During processing, a dependency relationship is established between the current text fragment and the referenced memory entries. Through these techniques, each representative agent can access historical information when processing fragments, thereby reducing semantic misunderstandings or redundant information extraction caused by missing context.

[0101] In one alternative embodiment, the process includes: each representative agent submits a semantic index of the currently processed segment to a context-aware memory pool; automatically matching memory entries associated with the semantic index; and performing joint analysis or reasoning by combining the segment with the matching memory entries to generate a processing result.

[0102] In one embodiment of this disclosure, each memory entry includes at least one of the following: the agent's identifier, the identifier of the memory entry, the processed text or fragment, a list of identifiers of memory entries on which the processed text or fragment depends, the processing result, and the confidence score of the processing result, the survival time, and the generation timestamp.

[0103] As an example, the timestamp for the generation of the processing result is the time when the processing result was generated.

[0104] As an example, the lifetime of the processing result is the time elapsed since the processing result was generated.

[0105] As an example, the list contains identifiers when the text or fragment being processed depends on multiple memory entries.

[0106] This embodiment constitutes a structured storage unit that supports contextual traceability, trust assessment, and timeliness management.

[0107] Each memory entry The fields included are as follows:

[0108]

[0109] Table 1

[0110] When the intelligent agent Generate new results At that time, in order to update the context-aware memory pool, the following is executed:

[0111] entry = {

[0112] "content": r_i,

[0113] "agent_id": A_i.id,

[0114] "timestamp": now(),

[0115] "confidence": c_i,

[0116] "dependencies": [m_j.id for m_j in referenced_entries],

[0117] "ttl": compute_ttl(c_i) # For example: ttl =10000 c_i

[0118] }

[0119] CAMP.write(entry).

[0120] In one embodiment of this disclosure, the context-aware memory pool provides a multi-dimensional query interface, which includes at least one of the following dimensions: keyword-based query, time window-based query, and confidence score-based query.

[0121] As an example, the query interface is the access channel provided by the context-aware memory pool, which includes queries of multiple dimensions.

[0122] In this embodiment, the context-aware memory pool provides a multi-dimensional query interface, including at least one of keyword-based queries, time-window-based queries, and confidence-score-based queries. This allows the agent to retrieve relevant memory entries based on semantic content, time range, or confidence level according to task requirements. Through these techniques, the agent can flexibly acquire historical information matching its needs when processing text fragments.

[0123] In one optional embodiment, the memory entries queried by each representative agent are sorted according to their confidence scores, and the memory entries returned to each representative agent have a lifespan that does not exceed the effective lifespan.

[0124] In one embodiment of this disclosure, memory entries with a lifetime greater than the effective lifetime are marked as invalid in the context-aware memory pool; memory entries with a confidence score less than a preset score are also marked as invalid in the context-aware memory pool.

[0125] The above-mentioned technical means ensure the timeliness and reliability of information in the context-aware memory pool.

[0126] Figure 4 This invention discloses a flowchart of a confidence score determination method according to an embodiment of the present disclosure. The method is as follows: Figure 4 As shown, it includes the following steps:

[0127] S401, Obtain user input information and determine the semantic matching degree between the processing results of each representative agent and the user input information;

[0128] S402, determine the source type of the fragment processed by each agent, and based on the source type of the fragment processed by each agent, determine the data source traceability strength of the processing results of each representative agent;

[0129] S403, based on the confidence score of the processing results of each representative agent, the data source strength, and the semantic matching degree corresponding to the processing results of each representative agent, determine the confidence score of the processing results of each representative agent.

[0130] As an example, user input can include user questions, comments, confirmation messages, or related queries.

[0131] As an example, semantic matching degree is used to measure the semantic similarity between the processing results generated by the representative agent and the user input information, reflecting the degree to which the output content matches the user's intent. The BERT model can be used to determine the semantic matching degree between the processing results of each representative agent and the user input information.

[0132] The exemplary source type describes the relationship between the fragments processed by each agent and the text to be processed.

[0133] As an example, data traceability strength reflects the reliability of the information sources on which the processing results depend, with each source type corresponding to a data traceability strength.

[0134] In this embodiment, the semantic matching degree between the processing results of each representative agent and the user input information is calculated, the source type of the text fragments processed by each agent is identified, the data source strength of the corresponding processing result is determined according to the source type, and the final confidence score of the processing result is calculated by combining the confidence score of the processing result itself, the data source strength, and the semantic matching degree. Through the above technical means, user feedback, data source characteristics, and semantic consistency can be integrated to generate a more comprehensive confidence assessment result.

[0135] The confidence score is calculated using the following formula, where the i-th value represents the agent's processing result. :

[0136] (4);

[0137] in, Let represent the confidence score of the processing result of the i-th agent. Let be the semantic matching degree corresponding to the processing result of agent i. Let be the data traceability strength of the processing result of agent i. , , This is a coefficient, which can be set by yourself, for example, set to... , .

[0138] Each token, when generated, has a corresponding softmax maximum probability value; therefore, the processing result for the agent... The formula for calculating its overall credibility is as follows:

[0139] (5);

[0140] in, This represents the softmax maximum probability value of the t-th token, where t is the token index and T is the maximum probability value. The number of tokens in the system.

[0141] The BERT model can be used for calculation:

[0142] (6);

[0143] The definitions of source and data traceability strength are shown in the table below:

[0144]

[0145] Table 2

[0146] Figure 5 This diagram illustrates a flowchart of a fusion priority determination method according to an embodiment of the present disclosure. The method is as follows: Figure 5 As shown, it includes the following steps:

[0147] S501, obtain the generation timestamp of the processing results of each representative agent;

[0148] S502, based on the confidence score and generation timestamp of the processing results of each representative agent, determine the fusion priority of the processing results of each representative agent.

[0149] As an example, the higher the confidence score of a processing result, the higher the fusion priority of that processing result; the closer the generation timestamp of a processing result is to the current time, the higher the fusion priority of that processing result.

[0150] In this embodiment, the timeliness and reliability of each processing result are comprehensively evaluated by combining the generation timestamp and confidence score, thereby determining the fusion priority of the processing results. Through the above technical means, both the credibility and timeliness of the results can be considered during information fusion.

[0151] The fusion priority of the processing results representing the i-th agent is calculated using the following formula. :

[0152] (7);

[0153] in, This represents the confidence score of the processing result of the i-th agent. This represents the timestamp of the generation of the i-th processing result representing the intelligent agent. In this embodiment, Represents the weighting coefficient, for example It is 0.6. It is the minimum value among the generation timestamps of all processing results representing intelligent agents. It is the maximum value among the generation timestamps of the processing results representing all agents.

[0154] In one embodiment of this disclosure, the processing results of each representative agent are merged based on the fusion priority of their processing results to obtain the parsing result of the text to be processed. This includes: performing conflict detection and / or consistency verification on the processing results of any two representative agents; when the processing results of any two representative agents pass the conflict detection and / or consistency verification, the processing results of each representative agent are merged based on the fusion priority of their processing results to obtain the parsing result.

[0155] In this embodiment, if all processing results involved in the merging process pass the conflict detection and / or consistency check, they are integrated according to their merging priority order to ultimately generate the parsed result of the text to be processed. If there are processing results that fail the conflict detection and / or consistency check, the processing results with lower merging priority can be deleted during the merging process. Through the above technical means, it is possible to effectively identify whether conflicts exist and whether consistency is satisfied before merging.

[0156] In one embodiment of this disclosure, conflict detection is performed on the processing results of any two representative agents, including: detecting whether the difference in confidence scores of the processing results of any two representative agents is greater than a preset threshold; if it is greater, then it is determined that the processing results of the two representative agents fail the conflict detection; or detecting whether the difference in generation timestamps of the processing results of any two representative agents is less than the larger of the effective lifetimes of the processing results of the two representative agents; if it is less, then it is determined that the processing results of the two representative agents fail the conflict detection.

[0157] The exemplary effective lifespan is used to determine whether a processing result has expired.

[0158] In this embodiment, the difference in confidence scores between any two processing results is checked to see if it exceeds a preset threshold. If it does, the two results fail the conflict detection, indicating that at least one result has significantly low credibility. The difference in generation timestamps between any two processing results is also checked to see if it is less than the larger of their effective lifespans. If it is less, the two results fail the conflict detection, indicating that they overlap in timeliness. By combining the confidence difference and the time validity window, the system can identify result pairs with unbalanced credibility or temporal redundancy, preventing low-quality or duplicate information from entering the fusion stage.

[0159] The processing result of the i-th agent is detected according to the following formula. and the processing result of the j-th agent. Conflict detection exists:

[0160] (8);

[0161] in, For the preset threshold, and They represent and Confidence score, It means "or". for and The absolute value of the difference between the generated timestamps. and They represent and The effective lifetime of the memory pool is stored in the context-aware memory pool. After this time, the result will be considered "expired" and will no longer be involved in merging or conflict detection.

[0162] In formula (8), if detected or detected If the condition is true, then the conflict detection is denied. Otherwise, the conflict detection is passed.

[0163] In one embodiment of this disclosure, detecting whether the difference between the generation timestamps of the processing results of any two representative agents is less than the larger of the effective lifetimes of the processing results of the two representative agents includes: determining the effective lifetime of the processing results of each representative agent based on the confidence score and content length of the processing results of each representative agent.

[0164] In this embodiment, the effective lifetime of each processing result is calculated based on its confidence score and content length, with higher confidence scores and longer content results assigned longer effective lifetimes. The larger of the effective lifetimes of two processing results is used as the benchmark for determining time overlap, and is employed for comparing the timestamp differences mentioned above. Through these techniques, the time validity window can be dynamically adjusted based on the quality and information density of the processing results.

[0165] Figure 6 A flowchart illustrating a method for determining effective survival time according to an embodiment of this disclosure is shown. The method is as follows: Figure 6 As shown, it includes the following steps:

[0166] S601, determine the basic validity period, credibility decay adjustment factor and content length gain coefficient;

[0167] S602 determines the effective lifespan of the processing results of each representative agent based on the basic validity period, the confidence decay adjustment factor, the content length gain coefficient, the confidence score and content length of the processing results of each representative agent.

[0168] As an example, the basic validity period is the initial validity period set for the processing results.

[0169] As an example, the confidence decay adjustment factor is a parameter that reflects the adjustment of confidence to effective survival time.

[0170] As an example, the content length gain coefficient is a coefficient that reflects the effect of content information density on effective lifetime.

[0171] In this embodiment, the final effective lifespan of a processing result is calculated by weighting the base validity period, the credibility decay adjustment factor, the content length gain coefficient, and the confidence score and content length of each processing result. Through these technical means, differentiated validity periods can be dynamically generated by comprehensively considering the base duration, credibility, and information density, thereby more accurately managing the lifecycle of contextual information.

[0172] In one exemplary embodiment, the following formula is used for calculation. Effective survival time :

[0173] (9);

[0174] in, This indicates the basic validity period, representing the basic lifespan of all processing results, for example, set to 60 seconds. is the credibility decay adjustment factor, and is a constant. This is the content length gain coefficient, which is a constant. for The length of the content can be used to determine the processing result. The number of tokens in the middle represents the information density.

[0175] In one embodiment of this disclosure, the consistency check of the processing results of any two representative agents includes: detecting whether there is a semantic conflict in the processing results of any two representative agents; if so, determining that the processing results of the two representative agents do not pass the consistency check.

[0176] In this embodiment, semantic analysis is used to compare the content involving the same entities, events, or attributes in two processing results to detect semantic conflicts, such as contradictory descriptions of the same fact, reversed chronological order, or logically mutually exclusive statements. If at least one semantic conflict is detected, the two processing results are determined to fail the consistency check. Through these technical means, inconsistent outputs at the logical or factual level can be identified before information fusion, thereby preventing erroneous or contradictory information from being merged to generate incorrect parsing results.

[0177] In one exemplary embodiment, consistency verification is performed according to the following formula:

[0178] (10);

[0179] in, This represents the set of all processing results representing intelligent agents. , A Boolean function indicating whether there is a contradiction, used to determine... and Is there a semantic contradiction? This represents the AND logic, performing an AND operation on all results. It indicates logical negation (not), meaning there is no contradiction.

[0180] In one embodiment of this disclosure, the method further includes: if the processing results of any two representative agents fail the consistency check, then backtracking the memory entries on which the segments processed by the two representative agents depend, in order to locate and correct the source of the error.

[0181] In this embodiment, based on the dependencies recorded in the two processing results, the memory entries referenced by each are located; these memory entries are compared and analyzed to identify memory entries or erroneous reasoning paths that cause semantic conflicts; the source of the error is marked as suspicious or invalid, and the corresponding agent is triggered to reprocess the relevant segments to correct the output results. Through the above technical means, it is possible to trace back from the result conflict to the source of information, thereby realizing the location and correction of erroneous information.

[0182] In an optional embodiment, the method further includes: when the processing results of any two representative agents fail the consistency check, obtaining a list of memory entry identifiers that each of the two processing results depends on, and tracing upwards layer by layer along the dependency chain displayed in the memory entry identifier list to the final memory entry; re-evaluating the credibility of the traced memory entry, and if it is found that the confidence score of a memory entry is lower than a set threshold or its lifetime has exceeded the effective lifetime, then marking it as invalid; reprocessing the relevant fragments based on the updated memory entry, and generating a corrected processing result. Through the above technical means, a closed-loop error correction mechanism based on dependency chains can be realized.

[0183] Figure 7 This invention discloses a flowchart of another dynamic consensus group partitioning method in an embodiment of the present disclosure, which is as follows: Figure 7 As shown, it includes the following steps:

[0184] S701, obtain the task descriptions of each agent;

[0185] S702, semantic vectorize the task description of each agent to obtain the semantic features corresponding to each agent;

[0186] S703, calculate the similarity matrix of semantic features corresponding to any two agents;

[0187] S704 uses a clustering algorithm to divide multiple agents based on the similarity matrix of semantic features corresponding to any two agents, forming a dynamic consensus group.

[0188] Consensus group division example 1:

[0189] Consensus Group G1: A, D;

[0190] Members: Agents A and D;

[0191] Group leader: A.

[0192] Consensus group division example 1:

[0193] Consensus Group G2: B, C;

[0194] Members: Agents B and C;

[0195] Group leader: B.

[0196] Figure 8 A schematic diagram illustrating memory entry details in an embodiment of this disclosure is shown, such as... Figure 8 As shown:

[0197] "Intermediate Inference v4.5" is generated based on "Original Data v4.3", and memory entry details are generated based on "Intermediate Inference v4.5". "Original Data v4.3" is a dependency of "Intermediate Inference v4.5".

[0198] The memory entry details table lists each field and its example values:

[0199] "content" means "broadband internet access", indicating the content of the entry;

[0200] The "agent_id" is "A1", indicating that it was generated by agent A1;

[0201] The "timestamp" value is 0.88, which records the generation timestamp;

[0202] The example value for “confidence” is “v14:20:05.123”, which reflects the confidence score.

[0203] The "dependencies" list contains [v4.3, v4.5], which shows the dependencies of this memory entry;

[0204] Finally, the "ttl" value of 500ms indicates the lifetime of this entry.

[0205] Therefore, the embodiments disclosed herein possess causal tracking capability and adaptive memory mechanism.

[0206] Figure 9 A flowchart of another text processing method in an embodiment of this disclosure is shown, the method as follows: Figure 9 As shown, it includes the following steps:

[0207] S901: Obtain the processing results for each item , …

[0208] S902: Calculate the fusion priority of each processing result.

[0209] S903: Detects for potential conflicts.

[0210] S904: A conflict exists. A reminder is issued to prompt the user to make a decision.

[0211] S905: No conflicts exist; all processing results are automatically merged, with the highest priority result retained during merging.

[0212] S906: Output the processing results after resolving the conflict.

[0213] S907: Perform consistency verification on each processing result.

[0214] S908: Consistency check passed. Merge all processing results and output the parsed result.

[0215] Based on the same inventive concept, this disclosure also provides a text processing apparatus, as shown in the following embodiments. Since the principle by which the text processing apparatus solves the problem is similar to that of the above method embodiments, the implementation of the text processing apparatus can refer to the implementation of the above method embodiments, and repeated details will not be elaborated further.

[0216] Figure 10 This disclosure illustrates a text processing apparatus, such as... Figure 10 As shown, the text processing device may include:

[0217] The processing unit 1001 is configured to distribute the multiple segments obtained by dividing the text to be processed to the representative agents of each dynamic consensus group for processing, and obtain the processing results of each representative agent.

[0218] The first determining unit 1002 is configured to determine the confidence score of the processing results of each representative agent;

[0219] The second determining unit 1003 is configured to determine the fusion priority of the processing results of each representative agent based on the confidence score of the processing results of each representative agent.

[0220] The merging unit 1004 is configured to merge the processing results of each representative agent based on the fusion priority of the processing results of each representative agent to obtain the parsing result of the text to be processed.

[0221] In some embodiments, the processing unit 1001 is further configured to extract semantic features of the task description of each agent; determine the semantic distance between the semantic features corresponding to any two agents; based on the semantic distance between the semantic features corresponding to any two agents, use a clustering algorithm to divide the multiple agents into multiple dynamic consensus groups, and determine the representative agent of the dynamic consensus group from one or more agents contained in each dynamic consensus group.

[0222] In some embodiments, the processing unit 1001 is further configured to query the context-aware memory pool by each representative agent to obtain the memory entries on which the fragments processed by each representative agent depend; process the fragments by each representative agent based on the queried memory entries to obtain the processing results of each representative agent; and generate the dependency relationship between the fragments processed by each representative agent and the memory entries on which the fragments depend.

[0223] In some embodiments, the first determining unit 1002 is further configured to acquire user input information, determine the semantic matching degree between the processing results of each representative agent and the user input information; determine the source type of the fragment processed by each agent, and determine the data source strength of the processing results of each representative agent based on the source type of the fragment processed by each agent; and determine the confidence score of the processing results of each representative agent based on the confidence score and data source strength of the processing results of each representative agent and the semantic matching degree corresponding to the processing results of each representative agent.

[0224] In some embodiments, the second determining unit 1003 is further configured to obtain the generation timestamp of the processing results of each representative agent; and determine the fusion priority of the processing results of each representative agent based on the confidence score and generation timestamp of the processing results of each representative agent.

[0225] In some embodiments, the merging unit 1004 is further configured to perform conflict detection and / or consistency verification on the processing results of any two representative agents; when the processing results of any two representative agents pass the conflict detection and / or consistency verification, the processing results of each representative agent are merged based on the fusion priority of the processing results of each representative agent to obtain the parsing result.

[0226] In some embodiments, the merging unit 1004 is further configured to detect whether the difference in confidence scores of the processing results of any two representative agents is greater than a preset threshold. If it is greater, the processing results of the two representative agents are determined to fail the conflict detection. Alternatively, it may detect whether the difference in generation timestamps of the processing results of any two representative agents is less than the larger of the effective lifetimes of the processing results of the two representative agents. If it is less, the processing results of the two representative agents are determined to fail the conflict detection.

[0227] In some embodiments, the merging unit 1004 is further configured to determine the effective lifetime of the processing results of each representative agent based on the confidence score and content length of the processing results of each representative agent.

[0228] In some embodiments, the merging unit 1004 is further configured to determine the base validity period, the credibility decay adjustment factor, and the content length gain coefficient; and to determine the effective lifespan of the processing results of each representative agent based on the base validity period, the credibility decay adjustment factor, the content length gain coefficient, the confidence score of the processing results of each representative agent, and the content length.

[0229] In some embodiments, the merging unit 1004 is further configured to detect whether there is a semantic conflict in the processing results of any two representative agents, and if so, to determine that the processing results of the two representative agents do not pass the consistency check.

[0230] In some embodiments, the merging unit 1004 is further configured to backtrack the memory entries on which the segments processed by the two representative agents depend if the processing results of any two representative agents fail the consistency check, in order to locate and correct the source of the error.

[0231] In some embodiments, the processing unit 1001 is further configured to query the context-aware memory pool through each representative agent to obtain the historical processing information of each representative agent, wherein the historical processing information of each agent includes multiple memory entries; process segments based on the historical processing information of each representative agent to obtain the processing results of each representative agent, and generate the dependency relationship between the segment processed by each representative agent and the memory entries on which the segment depends.

[0232] In one embodiment of this disclosure, each memory entry includes at least one of the following: the agent's identifier, the identifier of the memory entry, the processed text or fragment, a list of identifiers of memory entries on which the processed text or fragment depends, the processing result, and the confidence score of the processing result, the survival time, and the generation timestamp.

[0233] In one embodiment of this disclosure, the context-aware memory pool provides a multi-dimensional query interface, which includes at least one of the following dimensions: keyword-based query, time window-based query, and confidence score-based query.

[0234] In some embodiments, the processing unit 1001 is further configured to mark memory entries with a lifetime greater than the effective lifetime as invalid in the context-aware memory pool; and to mark memory entries with a confidence score less than a preset score as invalid in the context-aware memory pool.

[0235] Those skilled in the art will understand that various aspects of this disclosure can be implemented as a system, method, or program product. Therefore, various aspects of this disclosure can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software aspects, collectively referred to herein as a "circuit," "module," or "system."

[0236] The following reference Figure 11 To describe an electronic device 1100 according to such an embodiment of the present disclosure. Figure 11The electronic device 1100 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments disclosed herein.

[0237] like Figure 11 As shown, the electronic device 1100 is presented in the form of a general-purpose computing device. The components of the electronic device 1100 may include, but are not limited to: at least one processor 1110, at least one memory 1120, and a bus 1130 connecting different system components (including memory 1120 and processor 1110).

[0238] The memory stores program code that can be executed by the processor 1110, causing the processor 1110 to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure. For example, the processor 1110 can perform the following steps of the above method embodiments: assigning multiple segments obtained by dividing the text to be processed to representative agents of each dynamic consensus group for processing, obtaining the processing results of each representative agent; determining the confidence score of the processing results of each representative agent; determining the fusion priority of the processing results of each representative agent based on the confidence score of the processing results of each representative agent; merging the processing results of each representative agent based on the fusion priority of the processing results of each representative agent to obtain the parsing result of the text to be processed.

[0239] The memory 1120 may include a readable medium in the form of volatile memory, such as random access memory (RAM) 11201 and / or cache memory 11202, and may further include read-only memory (ROM) 11203.

[0240] The memory 1120 may also include a program / utility 11204 having a set (at least one) of program modules 11205, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0241] Bus 1130 can represent one or more of several types of bus structures, including a memory bus or memory controller, peripheral bus, graphics acceleration port, processor, or a local bus using any of the various bus structures.

[0242] Electronic device 1100 can also communicate with one or more external devices 1140 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more text processing devices that enable user interaction with electronic device 1100, and / or with any device that enables electronic device 1100 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 1150. Furthermore, electronic device 1100 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 1160. As shown, network adapter 1160 communicates with other modules of electronic device 1100 via bus 1130. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 1100, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0243] In the disclosed exemplary embodiments, a computer-readable storage medium is also provided, which may be a readable signal medium or a readable storage medium.

[0244] In some possible implementations, various aspects of this disclosure may also be implemented as a program product comprising program code that, when run on a terminal device, causes the terminal device to perform the steps described in the foregoing “Detailed Description” section of this specification according to various exemplary embodiments of this disclosure.

[0245] More specific examples of computer-readable storage media in this disclosure may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, 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 devices, magnetic storage devices, or any suitable combination of the foregoing.

[0246] In this disclosure, a computer-readable storage medium may include a data signal propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of transmitting, propagating, or transmitting a program for use by or in connection with an instruction execution system, apparatus, or device.

[0247] Optionally, the program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0248] In practical implementation, program code for performing the operations of this disclosure can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on a terminal device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0249] This disclosure provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform a text processing method provided in various alternative embodiments of this disclosure.

[0250] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0251] Furthermore, although the steps of the method in this disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0252] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, mobile terminal, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0253] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope of this disclosure is indicated by the appended claims.

Claims

1. A text processing method characterized by, include: The multiple segments obtained from dividing the text to be processed are assigned to representative agents of each dynamic consensus group for processing, and the processing results of each representative agent are obtained. Each dynamic consensus group is obtained by dividing agents using a clustering algorithm. Each dynamic consensus group includes one or more agents. The representative agent of each dynamic consensus group is the agent determined by the clustering algorithm as the group leader of the dynamic consensus group. The processing result of each representative agent is the output generated by the representative agent after performing semantic analysis on the assigned segments. Determine the confidence score of the processing results for each representative agent; Based on the confidence scores of the processing results of each representative agent, the fusion priority of the processing results of each representative agent is determined. Based on the fusion priority of the processing results of each representative agent, the processing results of each representative agent are merged to obtain the parsing result of the text to be processed. The step of merging the processing results of each representative agent based on the fusion priority of their processing results to obtain the parsing result of the text to be processed includes: Perform conflict detection and / or consistency verification on the processing results of any two representative agents; When the processing results of any two representative agents both pass the conflict detection and / or the consistency verification, the processing results of each representative agent are merged based on the fusion priority of their processing results to obtain the parsing result.

2. The method of claim 1, wherein, Before assigning the multiple segments obtained from dividing the text to be processed to representative agents of each dynamic consensus group for processing, and obtaining the processing results of each representative agent, the method further includes: Extract semantic features from the task descriptions of each agent; Determine the semantic distance between the semantic features corresponding to any two agents; Based on the semantic distance between the semantic features of any two agents, a clustering algorithm is used to divide multiple agents into multiple dynamic consensus groups, and the representative agent of each dynamic consensus group is determined from one or more agents contained in each dynamic consensus group.

3. The method according to claim 1, characterized in that, The process of assigning multiple segments of the text to be processed to representative agents of each dynamic consensus group for processing, and obtaining the processing results of each representative agent, includes: By querying the context-aware memory pool by each representative agent, the memory entries on which the fragments processed by each representative agent depend can be obtained. Each representative agent processes a fragment based on the queried memory entry, obtains the processing result of each representative agent, and generates the dependency relationship between the fragment processed by each representative agent and the memory entry on which the fragment depends.

4. The method according to claim 1, characterized in that, The determination of the confidence score of the processing results of each representative agent includes: Obtain user input information and determine the semantic matching degree between the processing results of each representative intelligent agent and the user input information; Determine the source type of the fragments processed by each agent, and based on the source type of the fragments processed by each agent, determine the data source strength of the processing results of each representative agent; Based on the confidence scores and data source strength of the processing results of each representative agent, as well as the semantic matching degree corresponding to the processing results of each representative agent, the confidence score of the processing results of each representative agent is determined.

5. The method according to claim 1, characterized in that, The process of determining the fusion priority of the processing results of each representative agent based on the confidence score of the processing results of each representative agent includes: Obtain the generation timestamp of the processing results of each representative agent; Based on the confidence scores and generation timestamps of the processing results of each representative agent, the fusion priority of the processing results of each representative agent is determined.

6. The method according to claim 1, characterized in that, The conflict detection of processing results for any two representative agents includes: The system checks whether the difference in confidence scores between the processing results of any two representative agents is greater than a preset threshold. If it is, the system determines that the processing results of the two representative agents fail the conflict detection; or... The difference between the generation timestamps of the processing results of any two representative agents is checked to see if it is less than the larger of the effective lifetimes of the processing results of the two representative agents. If it is less, it is determined that the processing results of the two representative agents do not pass the conflict detection.

7. The method according to claim 6, characterized in that, The step of detecting whether the difference between the generation timestamps of the processing results of any two representative agents is less than the larger of the effective lifetimes of the processing results of the two representative agents includes: Based on the confidence score and content length of the processing results of each representative agent, the effective lifespan of the processing results of each representative agent is determined.

8. The method according to claim 7, characterized in that, The determination of the effective lifetime of the processing results of each representative agent based on the confidence score and content length of the processing results of each representative agent includes: Determine the basic validity period, credibility decay adjustment factor, and content length gain coefficient; Based on the basic validity period, the confidence decay adjustment factor, the content length gain coefficient, and the confidence score and content length of the processing results of each representative agent, the effective survival time of the processing results of each representative agent is determined.

9. The method according to claim 1, characterized in that, The consistency check of the processing results of any two representative agents includes: If a semantic conflict exists between the processing results of any two representative agents, then it is determined that the processing results of the two representative agents do not pass the consistency check.

10. The method according to claim 9, characterized in that, The method further includes: If the processing results of any two representative agents fail the consistency check, the memory entries on which the segments processed by the two representative agents depend are traced back to locate and correct the source of the error.

11. The method according to claim 1, characterized in that, The process of assigning multiple segments of the text to be processed to representative agents of each dynamic consensus group for processing, and obtaining the processing results of each representative agent, includes: By querying the context-aware memory pool by each representative agent, the historical processing information of each representative agent is obtained. The historical processing information of each agent includes multiple memory entries. By processing fragments based on the historical processing information of each representative agent, the processing results of each representative agent are obtained, and the dependency relationship between the fragment processed by each representative agent and the memory entry on which the fragment depends is generated.

12. The method according to claim 3 or 11, characterized in that, Each memory entry includes at least one of the following: the agent's identifier, the identifier of the memory entry, the processed text or fragment, a list of identifiers of the memory entries on which the processed text or fragment depends, the processing result, and the confidence score of the processing result, the survival time, and the generation timestamp.

13. The method according to claim 3 or 11, characterized in that, The context-aware memory pool provides a multi-dimensional query interface, which includes at least one of the following dimensions: keyword-based query, time window-based query, and confidence score-based query.

14. The method according to claim 3 or 11, characterized in that, In the context-aware memory pool, memory entries with a lifespan greater than the effective lifespan are marked as invalid. In the context-aware memory pool, memory entries with confidence scores lower than a preset score are marked as invalid.

15. A text processing device, characterized in that, include: The processing unit is configured to assign multiple segments obtained from dividing the text to be processed to representative agents of each dynamic consensus group for processing, and obtain the processing results of each representative agent. Each dynamic consensus group is obtained by dividing agents using a clustering algorithm. Each dynamic consensus group includes one or more agents. The representative agent of each dynamic consensus group is the agent determined by the clustering algorithm as the group leader of the dynamic consensus group. The processing result of each representative agent is the output generated by the representative agent after performing semantic analysis on the assigned segments. The first determining unit is configured to determine the confidence score of the processing results of each representative agent; The second determining unit is configured to determine the fusion priority of the processing results of each representative agent based on the confidence score of the processing results of each representative agent. The merging unit is configured to merge the processing results of each representative agent based on the fusion priority of the processing results of each representative agent, so as to obtain the parsing result of the text to be processed. The merging unit is further configured to perform conflict detection and / or consistency verification on the processing results of any two representative agents; when the processing results of any two representative agents pass the conflict detection and / or consistency verification, the processing results of each representative agent are merged based on the fusion priority of the processing results of each representative agent to obtain the parsing result.

16. An electronic device, characterized in that, include: processor; as well as Memory for storing the executable instructions of the processor; The processor is configured to execute the method of any one of claims 1-14 by executing the executable instructions.

17. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1-14.

18. A computer program product comprising computer instructions stored in a computer-readable storage medium, wherein the computer instructions, when executed by a processor, implement the operation instructions of the method according to any one of claims 1-14.

Images