Semantic task processing method and apparatus, electronic device, and storage medium

Abstract

The application provides a semantic task processing method and device, electronic equipment and storage medium. The semantic task processing method comprises: acquiring language interaction information input by a user in a multi-round dialogue, analyzing the language interaction information, and generating a plurality of basic semantic tasks; generating a semantic task link based on a dependency relationship between the basic semantic tasks, and assigning a corresponding task intelligent agent to each basic semantic task based on the semantic task link; executing the corresponding basic semantic task through each task intelligent agent, generating a task signature of each basic semantic task, and generating a semantic link fingerprint based on the task signature. The application analyzes the language interaction information into basic semantic tasks, structures the basic semantic tasks into executable and traceable semantic task links, and finally generates a semantic link fingerprint, which guarantees the flexibility of the generated question and answer, and improves the controllability, reliability and traceability of complex multi-step queries.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a semantic task processing method, apparatus, electronic device, and storage medium. Background Technology

[0002] Existing large-scale model-driven Business Intelligence (BI) or ChatBI products mostly only support single-round instruction parsing and single-time Structured Query Language (SQL) query generation, making it difficult to support multi-step, multi-data-table, and multi-data-source combined queries required in complex business analysis scenarios. Furthermore, users may change or supplement their intentions multiple times during natural language interactions; traditional intelligent agent processing can lead to contextual confusion, inconsistent logical execution, and untraceable processes, hindering auditing. Summary of the Invention

[0003] This invention provides a semantic task processing method, apparatus, electronic device, and storage medium to address the deficiencies in the prior art.

[0004] This invention provides a semantic task processing method, comprising: The system acquires language interaction information input by the user in multi-turn dialogues, parses the language interaction information, and generates multiple basic semantic tasks. Based on the dependencies between the basic semantic tasks, a semantic task link is generated, and a corresponding task agent is assigned to each of the basic semantic tasks based on the semantic task link. Each task agent executes the corresponding basic semantic task, generates a task signature for each basic semantic task, and generates a semantic link fingerprint based on the task signature.

[0005] According to a semantic task processing method provided by the present invention, the step of assigning corresponding task agents to each of the basic semantic tasks based on the semantic task link includes: Based on the task attributes, priorities, data source distribution, and semantic task links of each basic semantic task, a corresponding task agent is assigned to each basic semantic task. The task agent includes one or more of the following agents: database query agent, data cleaning agent, indicator calculation agent, result verification agent, and anomaly interpretation agent.

[0006] According to a semantic task processing method provided by the present invention, the step of executing the corresponding basic semantic task through each task agent, generating a task signature for each basic semantic task, and generating a semantic link fingerprint based on the task signature includes: Each task agent executes the corresponding basic semantic task, and records task data during the execution of the basic semantic task; wherein, the task data includes the input, output, intermediate variables, and call data of the task agent; A hash operation is performed on the task data of each of the basic semantic tasks to generate the task signature corresponding to each of the basic semantic tasks; The semantic link fingerprint is obtained by connecting the signatures of each task in the semantic task link.

[0007] According to a semantic task processing method provided by the present invention, after each task agent executes the corresponding basic semantic task, generates a task signature for each basic semantic task, and generates a semantic link fingerprint based on the task signature, the method further includes: Obtain the task execution results of each of the basic semantic tasks, generate interactive feedback information based on the task execution results, and return the interactive feedback information to the user; Obtain the user's identity identifier, and archive the semantic link fingerprint, the user's identity identifier, and the interaction feedback information.

[0008] According to a semantic task processing method provided by the present invention, the step of generating semantic task links based on the dependencies between the basic semantic tasks includes: Using the basic semantic tasks as nodes and the dependencies between the basic semantic tasks as directed edges, a directed acyclic graph is constructed. The semantic task link is determined based on the directed acyclic graph.

[0009] According to a semantic task processing method provided by the present invention, after generating a semantic task link based on the dependency relationship between the basic semantic tasks, the method further includes: If a change operation to the semantic task link is detected, the semantic task to be changed is determined based on the change operation; Based on the changed semantic task, a task-affected sub-link is determined in the semantic task link, and the changed semantic task is updated in the task-affected sub-link.

[0010] According to a semantic task processing method provided by the present invention, after each task agent executes the corresponding basic semantic task, generates a task signature for each basic semantic task, and generates a semantic link fingerprint based on the task signature, the method further includes: A process template is generated based on the semantic task link so that when new language interaction information is received, the new language interaction information is processed based on the process template. The process template includes agent allocation information, data information, and dependency relationships.

[0011] The present invention also provides a semantic task processing apparatus, comprising: The parsing module is configured to acquire language interaction information input by the user in multi-turn dialogues, parse the language interaction information, and generate multiple basic semantic tasks; The allocation module is configured to generate semantic task links based on the dependencies between the basic semantic tasks, and to allocate corresponding task agents to each of the basic semantic tasks based on the semantic task links. The execution module is configured to execute the corresponding basic semantic task through each of the task agents, generate a task signature for each of the basic semantic tasks, and generate a semantic link fingerprint based on the task signature.

[0012] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory that can run on the processor, wherein the processor executes the computer program to implement the semantic task processing method as described above.

[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the semantic task processing method as described above.

[0014] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the semantic task processing method as described above.

[0015] The semantic task processing method, apparatus, electronic device, and storage medium provided by this invention parse language interaction information into basic semantic tasks, structure these basic semantic tasks into executable and traceable semantic task links, and have them collaboratively completed by a dedicated task intelligence agent. Finally, a unique semantic link fingerprint is generated. This greatly improves the controllability, reliability, and traceability of complex multi-step queries while ensuring the flexibility of generative question answering, providing a solid technical foundation for data analysis. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is one of the flowcharts of the semantic task processing method provided by the present invention.

[0018] Figure 2 This is the second flowchart of the semantic task processing method provided by the present invention.

[0019] Figure 3 This is a schematic diagram of the semantic task processing device provided by the present invention.

[0020] Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0022] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. The collection, use and processing of the relevant data shall comply with the relevant laws, regulations and standards of the relevant countries and regions. Corresponding operation portals are provided for users to choose to authorize or refuse.

[0023] Figure 1 This is a flowchart illustrating a semantic task processing method according to an exemplary embodiment. For example... Figure 1 As shown in an exemplary embodiment, the semantic task processing method includes steps 110 to 130, which are described in detail below.

[0024] Step 110: Obtain the language interaction information input by the user in the multi-turn dialogue, parse the language interaction information, and generate multiple basic semantic tasks.

[0025] In this embodiment of the invention, language interaction information can be understood as all the information input by the user through interaction with the system to achieve a certain data analysis or query purpose. This information can take various forms; for example, it can be text input, where the user enters "Query the sales figures for the first quarter in region xx and compare them with the same period last year" in the dialog box via the keyboard; it can also be voice input, where the system converts it into text using a speech recognition module. This interaction can be single-turn, where the user inputs the complete query command at once; more generally, the interaction is multi-turn, where the user gradually supplements or corrects the context in multiple rounds of dialogue. For example, the user might first input "Query the sales figures for the first quarter," and after the system returns the results, the user might add "limited to region xx," and then input "compare with the same period last year." During the dialogue, a multi-turn context tracking mechanism is used to determine the degree of dependency between the user's input and historical links in real time, thereby ensuring that similar commands can be mapped to the same semantic link, rather than being repeatedly generated.

[0026] After acquiring the language interaction information, it needs to be parsed. The goal of the parsing process is to generate multiple basic semantic tasks. These basic semantic tasks are defined as indivisible atomic units of operation with clearly defined execution goals that constitute a user's complete intent. The parsing process goes beyond traditional intent recognition and slot filling; it involves deeper semantic understanding and task decomposition. During parsing, the upstream and downstream dependencies between these basic semantic tasks, their parallel execution capabilities, and their data source call requirements are automatically marked. The parsing result outputs structured data, laying a visual foundation for subsequent multi-role scheduling.

[0027] In one specific implementation, parsing may include: Intent recognition: Identify the user's primary purpose, such as querying, comparing, summing, predicting, etc.

[0028] Semantic chunking and dependency determination: Combining syntactic analysis (e.g., constructing a syntactic tree) and contextual memory, a complex long sentence or multi-turn dialogue content is broken down into multiple logically independent semantic chunks. For example, "query sales figures for East China in the first quarter" can be broken down into semantic chunks such as "limited to the first quarter", "limited to East China", and "calculate sales figures". Simultaneously, the logical dependencies between these semantic chunks are determined.

[0029] Task unit generation: Each semantic block is transformed into a standardized basic semantic task. For example, calculating sales revenue is transformed into a data aggregation task, and limiting the region to East China is transformed into a data filtering task. Each basic semantic task contains all the elements required to perform the task, such as operation type, parameters, and objects.

[0030] For example, if a user's language interaction request is to query the total sales of product A in locations xx and yy for the first quarter of 2024, and to display the trend by month, parsing it might generate the following basic semantic tasks: Task 1: The time frame for screening is from January 1, 2024 to March 31, 2024; Task 2: Select product A; Task 3: Filter regions as xx or yy; Task 4: Group the filtered data by month; Task 5: Calculate the sum of sales revenue for each group of data; Task 6: Present the calculation results in the form of a trend chart.

[0031] Step 120: Generate semantic task links based on the dependencies between the basic semantic tasks, and assign corresponding task agents to each of the basic semantic tasks based on the semantic task links.

[0032] In this embodiment of the invention, after generating multiple discrete basic semantic tasks, they need to be organized into an ordered, executable workflow; this is the process of generating a semantic task chain. In this embodiment, a semantic task chain refers to a logical structure that describes all basic semantic tasks, their execution order, and data flow relationships. The chain is generated based on the dependencies resolved in the previous step. For example, filtering (tasks 1, 2, 3) must be performed first before grouping and calculation (tasks 4, 5), and finally, display (task 6).

[0033] In one embodiment, generating a semantic task chain can be achieved by organizing all the parsed basic semantic tasks into an ordered list or a more complex topology according to their logical sequence and data transmission relationships.

[0034] Subsequently, task agents are assigned to each basic semantic task based on the semantic task chain. A task agent can be understood as a software agent or functional module with specific functions, capable of autonomously or semi-autonomously executing a specific type of basic semantic task. The assignment process involves assigning an executor to each task step to be executed. In this embodiment of the invention, task agent pools with different functions can be preset, and matching and assignment can be performed according to the type of basic semantic task and the semantic task chain. For example, data filtering tasks are assigned to filtering agents, and data computation tasks are assigned to computation agents.

[0035] Step 130: Each task agent executes the corresponding basic semantic task to generate a task signature for each basic semantic task, and generates a semantic link fingerprint based on the task signature.

[0036] In this embodiment of the invention, after allocation, each task agent begins to execute its corresponding basic semantic task. For example, an agent responsible for data retrieval may generate and execute an SQL statement according to task requirements, while an agent responsible for computation may perform aggregation operations on the data in memory.

[0037] During execution, task signatures are generated for each basic semantic task. A task signature can be understood as a snapshot or record of the execution behavior of a single basic semantic task, uniquely identifying this execution.

[0038] Finally, a semantic link fingerprint is generated based on the task signature. The semantic link fingerprint is a unique and verifiable identifier that represents the entire semantic task link from generation to completion.

[0039] The semantic task processing method provided in this embodiment structures natural language queries into executable and traceable semantic task links, which are then collaboratively completed by a dedicated task agent, ultimately generating a unique link fingerprint. This significantly improves the controllability, reliability, and traceability of complex multi-step queries while ensuring the flexibility of generative question answering, providing a solid technical foundation for enterprise-level intelligent data analysis.

[0040] In an exemplary embodiment of the present invention, the step of assigning corresponding task agents to each of the basic semantic tasks based on the semantic task link includes: Based on the task attributes, priorities, data source distribution, and semantic task links of each basic semantic task, a corresponding task agent is assigned to each basic semantic task; the task agent includes one or more of the following agents: database query agent, data cleaning agent, indicator calculation agent, result verification agent, and anomaly explanation agent.

[0041] In this embodiment of the invention, the allocation process can be completed by a multi-role intelligent agent scheduler. The multi-role intelligent agent scheduler analyzes the link structure of the semantic task chain and the information of the basic semantic tasks, and matches the most suitable intelligent agent for each basic semantic task. The multi-role intelligent agent scheduler consists of two parts: a role matching engine and an execution queue manager. The role matching engine intelligently determines whether to create a new intelligent agent or reuse an existing intelligent agent based on the context requirements of each basic semantic task, avoiding resource waste. The execution queue manager dynamically sorts the execution order according to system load and task urgency, supports task interruption, priority switching, and cross-role state synchronization, ensuring that the generative query process remains stable and controllable even under multi-user concurrency.

[0042] The multi-role agent scheduler considers a wider range of dimensions when making allocation decisions: Task attributes refer to the type of the basic semantic task. For example, is the basic semantic task data retrieval, data cleaning, indicator calculation, result verification, or anomaly interpretation?

[0043] Priority refers to the urgency or importance of a fundamental semantic task. For example, in a semantic task chain, a core data retrieval task, which is a prerequisite for multiple subsequent tasks, may have a higher priority than an auxiliary data formatting task. A multi-role intelligent agent scheduler can prioritize allocating computing resources to high-priority tasks.

[0044] Data source distribution refers to where the data to be processed by the basic semantic task is located. For example, if a task needs to connect to a company's internal MySQL database and a cloud-based Hive data warehouse, the multi-role agent scheduler needs to coordinate agents that can access these two data sources, or break the task down and assign it to different agents.

[0045] In this embodiment of the invention, the task agent includes, but is not limited to: Database query agents are responsible for understanding data query tasks, generating, optimizing, and executing structured query languages ​​(such as SQL, Cypher, etc.), and interacting with various databases or data warehouses to obtain raw data; The data cleaning agent is responsible for preprocessing the acquired raw data, such as handling missing values, removing duplicates, and performing format conversions, in order to ensure data quality. The indicator calculation agent is responsible for performing various complex business indicator calculations, such as year-on-year / month-on-month growth rate, moving average, user retention rate, etc. It may have a rich library of mathematical and statistical functions built in. The result verification agent is responsible for verifying the correctness and reasonableness of the output results of other agents. For example, it can check whether a calculated sales amount is within a reasonable range, or cross-validate whether data from two different sources are consistent. The anomaly explanation agent is activated when the result verification agent detects an anomaly, or when a user directly asks why sales have declined. This agent attempts to perform attribution analysis from multiple dimensions (such as time, region, product line, etc.) to find possible reasons for the anomaly.

[0046] By introducing an allocation mechanism based on task attributes, priorities, and data source distribution, and by adopting an intelligent agent system that includes multiple professional roles such as querying, cleaning, calculation, verification, and interpretation, this embodiment can achieve more accurate, efficient, and intelligent task scheduling, significantly improving the automation level of the entire data analysis process and the reliability of the final results.

[0047] In an exemplary embodiment of the present invention, the step of executing the corresponding basic semantic task through each of the task agents, generating a task signature for each of the basic semantic tasks, and generating a semantic link fingerprint based on the task signature includes: Each task agent executes the corresponding basic semantic task, and records task data during the execution of the basic semantic task; wherein, the task data includes the input, output, intermediate variables, and call data of the task agent; A hash operation is performed on the task data of each of the basic semantic tasks to generate the task signature corresponding to each of the basic semantic tasks; The semantic link fingerprint is obtained by connecting the signatures of each task in the semantic task link.

[0048] In this embodiment of the invention, task data is the raw material constituting the task signature, which meticulously records every detail of task execution. Specifically, task data includes: The input to the task agent is the original data or the output of the upstream basic semantic task that it relies on when performing the basic semantic task.

[0049] The output of the task agent: the result generated after the task is completed; Intermediate variables: Variables generated during task execution for temporary calculations or state saving; Call data: Records of calls to external services, function libraries, or models during task execution, including the name of the call, the parameters passed in, and the results returned; In this embodiment, the task signature is generated using cryptographic hash operations, such as the SHA-256 algorithm. Specifically, all task data for a given basic semantic task is serialized, for example, converted into a JSON string, and then a hash operation is performed on this serialized data. Due to the characteristics of hash functions, even a tiny change in the task data will cause a large and unpredictable change in the generated hash value. This ensures the uniqueness and tamper-proof nature of the task signature.

[0050] In this embodiment, a semantic link fingerprint is obtained by connecting the signatures of each task based on the semantic task link. This connection is not a simple string concatenation, but a chain structure forming a hash chain. Specifically, the task signature (hash_1) of the first basic semantic task serves as the starting point of the chain. When calculating the task signature of the second basic semantic task, its input includes not only its own task data but also the task signature hash_1 of the first task. Therefore, the signature hash_2 of the second task = Hash(task data_2 + hash_1). Similarly, the signature hash_n of the nth task = Hash(task data_n + hash_{n-1}). Finally, the signature hash_n of the last task in the semantic task link, since it indirectly contains information from all the preceding task signatures, can serve as the fingerprint of the entire semantic link.

[0051] By employing the aforementioned hash-based and chain-connected approach to generate task signatures and semantic link fingerprints, this embodiment provides an immutable and verifiable digital evidence for the entire data analysis process. This not only ensures the traceability of the results but also provides a high degree of credibility for subsequent auditing, compliance checks, and problem investigation.

[0052] In an exemplary embodiment of the present invention, after the step of executing the corresponding basic semantic task by each of the task agents, generating a task signature for each of the basic semantic tasks, and generating a semantic link fingerprint based on the task signature, the method further includes: Obtain the task execution results of each of the basic semantic tasks, generate interactive feedback information based on the task execution results, and return the interactive feedback information to the user; Obtain the user's identity identifier, and archive the semantic link fingerprint, the user's identity identifier, and the interaction feedback information.

[0053] In this embodiment of the invention, the meaningful task execution results generated in the semantic task chain are integrated. For example, the task execution result is a calculated total sales amount or a generated data chart. The obtained task execution result is transformed into interactive feedback information that is easy for the user to understand. The interactive feedback information can be an answer described in natural language, such as "The total sales amount of product A in the first quarter of 2024 was 'aa' million yuan". The interactive feedback information can also be a visual chart or report, or a data file that can be downloaded by the user. The obtained interactive feedback information is returned to the user through the user interface, completing the core objective of this query.

[0054] User identification is data that uniquely identifies the user currently performing the operation, such as a user's login name or employee ID. After a complete query interaction, all core evidence generated during that interaction is archived. The archived content mainly includes: Semantic link fingerprints represent a complete and tamper-proof record of what was done and how it was done in this query. The user identification identifies who did it.

[0055] Archiving can be done by writing this information to a dedicated, immutable chained log or audit database, along with an operation timestamp.

[0056] This embodiment improves the user experience and forms a complete interactive loop by adding result feedback and identity-related archiving steps. On the other hand, it binds each intelligent data analysis behavior to a specific user identity and records it in a highly reliable manner, which greatly enhances the system's security, auditability, and accountability. It is particularly suitable for enterprise environments with strict requirements for data security and operational compliance.

[0057] In this embodiment of the invention, it is also supported to write semantic links, role calls, execution results, and version information into an immutable hash log. Each link version, each task scheduling, and each role execution corresponds to a unique hash entry, and the entries are linked by timestamps and chain pointers to prevent malicious modification later. Fine-grained query and permission management interfaces are also provided, allowing administrators to trace query history by user, role, task unit, or data source to quickly locate abnormal operations or execution deviations. For data access scenarios requiring compliance with regulations, the chained log can export a complete execution report, providing one-stop traceability capabilities for cross-departmental compliance audits in conjunction with semantic link fingerprinting. Furthermore, this module supports triggering security alerts and automatically freezing relevant roles when multiple execution failures or abnormal drifts occur, preventing data leakage or resource abuse due to erroneous scheduling.

[0058] In an exemplary embodiment of the present invention, generating semantic task links based on the dependencies between the basic semantic tasks includes: Using the basic semantic tasks as nodes and the dependencies between the basic semantic tasks as directed edges, a directed acyclic graph is constructed. The semantic task link is determined based on the directed acyclic graph.

[0059] In this embodiment of the invention, basic semantic tasks are treated as nodes, and the dependencies between basic semantic tasks are treated as directed edges. A Directed Acyclic Graph (DAG) is constructed to determine the semantic task chain. The DAG is used to manage the dependency order, context passing, and multi-role assignment relationships of basic semantic tasks, ensuring that data flow is logically closed-loop and free from infinite loops.

[0060] Specifically, a directed edge represents a path from node A to node B if the execution of task B depends on the result of task A. This edge clearly expresses the dependencies between tasks and the flow of data.

[0061] Acyclic graphs indicate that there are no circular dependencies in the entire graph. For example, A depends on B, and B depends on A, which ensures the logical correctness and termination of the task flow.

[0062] By employing a directed acyclic graph as the concrete implementation of semantic task links, this embodiment can not only accurately and unambiguously express the dependencies between various subtasks in complex queries, but also provide powerful data structure support for subsequent execution optimization (such as parallel scheduling), link visualization, and dynamic reordering (such as inserting or deleting nodes in the graph). It is the core foundation for realizing an efficient, flexible, and interpretable data analysis process.

[0063] In this embodiment of the invention, visualization of the semantic task chain is also supported. Users can view with one click which data tables, roles, and verification steps were called in this query, and can even export it as an audit report. For multi-round queries, the semantic chain fingerprint can be used in conjunction with version control to automatically record the topological changes of the semantic task chain during its evolution, facilitating comparison or rollback of historical versions when needed.

[0064] In an exemplary embodiment of the present invention, after generating the semantic task chain based on the dependency relationship between the basic semantic tasks, the method further includes: If a change operation to the semantic task link is detected, the semantic task to be changed is determined based on the change operation; Based on the changed semantic task, a task-affected sub-link is determined in the semantic task link, and the changed semantic task is updated in the task-affected sub-link.

[0065] In this embodiment of the invention, a change operation refers to a user's intention to modify an established analysis process during a multi-turn dialogue. This operation is dynamic and immediate. For example, when preparing to execute or executing the query "Query sales figures for region xx in the first quarter," the user might input "Add the pp region" or "Change sales figures to profit." The system needs to be able to detect that this is a change operation, not a completely new query.

[0066] Based on the detected change operations, the specific change semantic task needs to be determined. For example, for "add pp region", the change semantic task is to update the parameter of the original region filtering task from xx to xx or pp. For "change sales revenue to profit", the change semantic task is to update the operation object of the aggregation calculation task from sales revenue to profit.

[0067] After determining the semantic task of the change, the system does not simply discard the entire link and start over. Instead, it can intelligently determine the sub-links affected by the task. The sub-links affected by the task refer to the subgraph in the semantic task link that starts from the node where the change occurs and extends to all subsequent nodes that ultimately depend on the output of that node.

[0068] For example, in the original semantic task chain, if a user modifies the region filtering task, then all subsequent tasks that depend on it, such as grouping, calculation, and display, are affected by the sub-chain. However, the time filtering task, which runs parallel to the region filtering task, is unaffected, and its already executed results can be retained.

[0069] Once the scope of impact is determined, all that is needed is to update the semantic changes in the affected sub-links and re-execute the tasks in those sub-links.

[0070] This embodiment introduces a dynamic response mechanism to changes, particularly by minimizing the update scope through computational tasks affecting sub-links. This allows users to iterate and refine their analytical approaches, much like conversing with a real analyst, while intelligently preserving existing computational results and avoiding unnecessary duplication of work, significantly improving the efficiency and user experience of exploratory data analysis.

[0071] In addition, this invention provides a link version control function, generating a unique version number and hash fingerprint for each link structure change, which is used for subsequent auditing and replayability. For complex scenarios, users can drag and drop to reorder links or insert branches through a graphical interface.

[0072] In an exemplary embodiment of the present invention, after the step of executing the corresponding basic semantic task by each of the task agents, generating a task signature for each of the basic semantic tasks, and generating a semantic link fingerprint based on the task signature, the method further includes: A process template is generated based on the semantic task link so that when new language interaction information is received, the new language interaction information is processed based on the process template. The process template includes agent allocation information, data information, and dependency relationships.

[0073] In this embodiment of the invention, after a complex query task is completed, the corresponding semantic task chain can be used as an analytical knowledge asset and saved as a process template. The process template will permanently store the core structure and information of the original semantic task chain, specifically including: Agent assignment information: records which type of agent performs each task step; Data information: Records the data sources, tables, fields, etc. involved in the task, but may parameterize specific values. For example, the template records the regions to be filtered, but the specific value "xx" can be used as a parameter to be filled in; Dependencies: The original DAG topology of the link is fully preserved, that is, the order and dependencies between tasks.

[0074] When another user or the same user subsequently initiates a new query, forming new language interaction information, such as entering "query the sales figures for the YY region in the second quarter", during the parsing phase, in addition to performing basic semantic decomposition, it will also be matched with the intent of the saved process template library.

[0075] If the query is found to closely match a process template named "Quarterly Sales Query by Region", that process template will be directly invoked, and the parameter slots in the process template will be filled with the specific parameters from the new query (second quarter, yy region). In this way, most of the complex semantic parsing and link construction process can be skipped, and a directly executable semantic task link can be quickly generated.

[0076] This embodiment achieves knowledge-based and reusable data analysis processes by introducing a process template generation and invocation mechanism. For common and highly repetitive analysis reports or query needs within enterprises, this mechanism can significantly reduce repetitive work and greatly improve analysis efficiency. Simultaneously, it can solidify expert analytical approaches into templates, lowering the barrier for new employees or non-professionals to use complex data analysis tools and promoting the overall improvement of data analysis capabilities within the organization.

[0077] In this embodiment of the invention, users can also edit the saved process template through the interface, such as splitting, merging, inserting new intelligent agents, or replacing data sources. It supports dynamically injecting new context variables, such as different time windows or business scenario parameters, before the process template is executed, thereby allowing a set of semantic links to be flexibly invoked in different scenarios, forming an upgrade closed loop from a single dialogue to a maintainable process.

[0078] This invention dynamically parses user language interaction information into multiple executable basic semantic tasks, which are then allocated, executed step-by-step, and context-aware by multi-role intelligent agents. This constructs an insertable, reprogrammable, and replayable multi-task generative data analysis workflow. Compared to traditional single-agent question-and-answer or static scripts, this invention supports task link fingerprinting for evidence storage, ensuring that data query and reasoning behaviors are verifiable and auditable throughout the entire process. This significantly improves the interpretability and reusability of results in complex analysis scenarios. Furthermore, through multi-role parallelism and dynamic priority scheduling, it reduces the impact of context drift on result accuracy, truly achieving the flexibility, controllability, and security of a generative BI system.

[0079] In one exemplary embodiment of the present invention, please refer to Figure 2 , Figure 2 This is a schematic diagram illustrating a semantic task processing method according to an exemplary embodiment, such as... Figure 2 As shown: Users input natural language, resulting in language interaction information. This information undergoes intent recognition and syntactic parsing to decompose tasks into units, generating multiple basic semantic tasks. Dependencies between these tasks are identified, and semantic task chains are generated based on these dependencies, forming a Directed Acyclic Graph (DAG) topology. A unique version number is generated for each generated semantic task chain, and this number is updated with each change to the chain structure. Simultaneously, fingerprint hashing is performed to generate semantic chain fingerprints. A multi-role scheduler is used to match and assign task agents to each basic semantic task. During execution queue management, the execution order of basic semantic tasks is dynamically sorted based on system load and task urgency, enabling concurrency and priority switching for these tasks.

[0080] Multi-role intelligent agents execute their assigned basic semantic tasks, recording input and output data while performing contextual tracing to record intermediate variables and call data. Based on the recorded input / output, intermediate variables, and call data, a semantic link fingerprint is generated. Throughout the process, users can regenerate or rearrange the basic semantic tasks. A workflow template is generated based on the generated semantic task link for reuse in subsequent multi-turn dialogues. When users need to backtrack to a specific step or view details, they can trace the execution record node by node through the semantic task link. Users can also export the complete or partial link as an audit report for cross-departmental approval or compliance checks, fully meeting enterprises' transparency requirements for generative BI activities.

[0081] The semantic task processing apparatus provided by the present invention will be described below. The semantic task processing apparatus described below can be referred to in correspondence with the semantic task processing method described above. It should be noted that the apparatus provided in the embodiments below and the method provided in the embodiments above belong to the same concept, and the specific way in which each module and unit performs operations has been described in detail in the method embodiments, and will not be repeated here.

[0082] In one exemplary embodiment of the present invention, please refer to Figure 3 , Figure 3 This is a semantic task processing apparatus according to an exemplary embodiment, comprising the following modules.

[0083] The parsing module 310 is configured to acquire language interaction information input by the user in a multi-turn dialogue, parse the language interaction information, and generate multiple basic semantic tasks. The allocation module 320 is configured to generate semantic task links based on the dependencies between the basic semantic tasks, and to allocate corresponding task agents to each of the basic semantic tasks based on the semantic task links. The execution module 330 is configured to execute the corresponding basic semantic task through each of the task agents, generate a task signature for each of the basic semantic tasks, and generate a semantic link fingerprint based on the task signature.

[0084] In an exemplary embodiment of the present invention, the allocation module 320 includes: The allocation submodule is configured to allocate the corresponding task agent to each of the basic semantic tasks based on the task attributes, priorities, data source distribution, and semantic task links of each of the basic semantic tasks. The task agent includes one or more of the following agents: database query agent, data cleaning agent, indicator calculation agent, result verification agent, and anomaly interpretation agent.

[0085] In an exemplary embodiment of the present invention, the execution module 330 includes: The execution submodule is configured to execute the corresponding basic semantic task through each of the task agents, and record task data during the execution of the basic semantic task; wherein, the task data includes the input, output, intermediate variables, and call data of the task agent; The hash operation submodule is configured to perform hash operations on the task data of each of the basic semantic tasks to generate the task signature corresponding to each of the basic semantic tasks; The connection submodule is configured to connect the signatures of each task based on the semantic task link to obtain the semantic link fingerprint.

[0086] In an exemplary embodiment of the present invention, the semantic task processing apparatus further includes: The acquisition module is configured to acquire the task execution results of each of the basic semantic tasks, generate interactive feedback information based on the task execution results, and return the interactive feedback information to the user. The archiving module is configured to obtain the user identity identifier and archive the semantic link fingerprint, the user identity identifier, and the interaction feedback information.

[0087] In an exemplary embodiment of the present invention, the allocation module 320 includes: A submodule is constructed, configured to use the basic semantic tasks as nodes and the dependencies between the basic semantic tasks as directed edges to construct a directed acyclic graph. The submodule is configured to determine the semantic task link based on the directed acyclic graph.

[0088] In an exemplary embodiment of the present invention, the semantic task processing apparatus further includes: The determination module is configured to determine the changed semantic task based on the detected change operation when a change operation to the semantic task link is detected. The update module is configured to determine the task-affected sub-links in the semantic task chain based on the changed semantic task, and update the changed semantic task in the task-affected sub-links.

[0089] In an exemplary embodiment of the present invention, the semantic task processing apparatus further includes: The generation module is configured to generate a process template based on the semantic task link, so that when new language interaction information is received, the new language interaction information is processed based on the process template; wherein, the process template includes agent allocation information, data information, and dependency relationships.

[0090] Figure 4An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4 As shown, the electronic device may include a processor 410, a communications interface 420, a memory 430, and a communication bus 440, wherein the processor 410, communications interface 420, and memory 430 communicate with each other via the communication bus 440. The processor 410 can call logical instructions in the memory 430 to execute a semantic task processing method, which includes: acquiring language interaction information input by the user in a multi-turn dialogue, parsing the language interaction information, and generating multiple basic semantic tasks; Based on the dependencies between the basic semantic tasks, a semantic task link is generated, and a corresponding task agent is assigned to each of the basic semantic tasks based on the semantic task link. Each task agent executes the corresponding basic semantic task, generates a task signature for each basic semantic task, and generates a semantic link fingerprint based on the task signature.

[0091] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0092] On the other hand, the present invention also provides a computer program product, the computer program product including a computer program, the computer program being able to be stored on a non-transitory computer-readable storage medium, the computer program being executed by a processor, the computer being able to execute the semantic task processing method provided by the above methods, the method including: acquiring language interaction information input by the user in a multi-turn dialogue, parsing the language interaction information, and generating multiple basic semantic tasks; Based on the dependencies between the basic semantic tasks, a semantic task link is generated, and a corresponding task agent is assigned to each of the basic semantic tasks based on the semantic task link. Each task agent executes the corresponding basic semantic task, generates a task signature for each basic semantic task, and generates a semantic link fingerprint based on the task signature.

[0093] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the semantic task processing method provided by the above methods, the method comprising: acquiring language interaction information input by a user in a multi-turn dialogue, parsing the language interaction information, and generating multiple basic semantic tasks; Based on the dependencies between the basic semantic tasks, a semantic task link is generated, and a corresponding task agent is assigned to each of the basic semantic tasks based on the semantic task link. Each task agent executes the corresponding basic semantic task, generates a task signature for each basic semantic task, and generates a semantic link fingerprint based on the task signature.

[0094] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0095] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A semantic task processing method, characterized in that, include: The system acquires language interaction information input by the user in multi-turn dialogues, parses the language interaction information, and generates multiple basic semantic tasks. Based on the dependencies between the basic semantic tasks, a semantic task link is generated, and a corresponding task agent is assigned to each of the basic semantic tasks based on the semantic task link. Each task agent executes the corresponding basic semantic task, generates a task signature for each basic semantic task, and generates a semantic link fingerprint based on the task signature.

2. The semantic task processing method according to claim 1, characterized in that, The process of assigning corresponding task agents to each of the basic semantic tasks based on the semantic task link includes: Based on the task attributes, priorities, data source distribution, and semantic task links of each basic semantic task, a corresponding task agent is assigned to each basic semantic task. The task agent includes one or more of the following agents: database query agent, data cleaning agent, indicator calculation agent, result verification agent, and anomaly interpretation agent.

3. The semantic task processing method according to claim 1, characterized in that, The step of executing the corresponding basic semantic task through each of the task agents, generating a task signature for each of the basic semantic tasks, and generating a semantic link fingerprint based on the task signature includes: Each task agent executes the corresponding basic semantic task, and records task data during the execution of the basic semantic task; wherein, the task data includes the input, output, intermediate variables, and call data of the task agent; A hash operation is performed on the task data of each of the basic semantic tasks to generate the task signature corresponding to each of the basic semantic tasks; The semantic link fingerprint is obtained by connecting the signatures of each task in the semantic task link.

4. The semantic task processing method according to claim 1, characterized in that, After the step of executing the corresponding basic semantic task through each task agent, generating a task signature for each basic semantic task, and generating a semantic link fingerprint based on the task signature, the method further includes: Obtain the task execution results of each of the basic semantic tasks, generate interactive feedback information based on the task execution results, and return the interactive feedback information to the user; Obtain the user's identity identifier, and archive the semantic link fingerprint, the user's identity identifier, and the interaction feedback information.

5. The semantic task processing method according to claim 1, characterized in that, The generation of semantic task links based on the dependencies between the basic semantic tasks includes: Using the basic semantic tasks as nodes and the dependencies between the basic semantic tasks as directed edges, a directed acyclic graph is constructed. The semantic task link is determined based on the directed acyclic graph.

6. The semantic task processing method according to any one of claims 1 to 5, characterized in that, After generating the semantic task chain based on the dependencies between the basic semantic tasks, the method further includes: If a change operation to the semantic task link is detected, the semantic task to be changed is determined based on the change operation; Based on the changed semantic task, a task-affected sub-link is determined in the semantic task link, and the changed semantic task is updated in the task-affected sub-link.

7. The semantic task processing method according to any one of claims 1 to 5, characterized in that, After the step of executing the corresponding basic semantic task through each task agent, generating a task signature for each basic semantic task, and generating a semantic link fingerprint based on the task signature, the method further includes: A process template is generated based on the semantic task link so that when new language interaction information is received, the new language interaction information is processed based on the process template. The process template includes agent allocation information, data information, and dependency relationships.

8. A semantic task processing device, characterized in that, include: The parsing module is configured to acquire language interaction information input by the user in multi-turn dialogues, parse the language interaction information, and generate multiple basic semantic tasks; The allocation module is configured to generate semantic task links based on the dependencies between the basic semantic tasks, and to allocate corresponding task agents to each of the basic semantic tasks based on the semantic task links. The execution module is configured to execute the corresponding basic semantic task through each of the task agents, generate a task signature for each of the basic semantic tasks, and generate a semantic link fingerprint based on the task signature.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the semantic task processing method as described in any one of claims 1 to 7.

10. A non-transitory 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 semantic task processing method as described in any one of claims 1 to 7.

Images