Insight analysis method based on observation platform and electronic device

Abstract

The invention discloses an observation platform-based insight analysis method and an electronic device, and relates to the technical field of artificial intelligence, and the method comprises the steps: carrying out the intention recognition of a received insight analysis request, and according to the recognized task intention, combining with a preset structured conversation analysis process configuration library, carrying out the analysis of the intention; generating an insight analysis execution path containing a multi-step analysis instruction; according to the task intention, intention keywords are extracted from the insight analysis request, semantic retrieval is carried out in a long-term memory vector library, and context information is obtained; and according to the context information, according to a multi-step analysis instruction in the insight analysis execution path, triggering calling of a target tool matched with the current step step by step, and taking a corresponding tool execution result as an input of a corresponding subsequent step to obtain an insight analysis result. According to the method and the device, the defects of poor semantic comprehension capability of a general agent and relatively poor SQL statement conversion precision are overcome, and the semantic comprehension capability of the agent and the data analysis efficiency are improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to an insight analysis method and electronic device based on an observation platform. Background Technology

[0002] With the rapid development and maturation of large language model technology, artificial intelligence is evolving from simple perceptual intelligence to cognitive and decision-making intelligence. Artificial Intelligence Agents (AI-Agents), as a core product of this trend, are becoming a crucial bridge connecting human intentions and the digital world. An AI-Agent is an intelligent entity with autonomous understanding, planning, execution, and feedback capabilities. It can receive natural language instructions from users, proactively break down tasks, invoke tools (such as APIs, databases, and code execution environments), and ultimately accomplish complex goals.

[0003] Currently, tracking the operation of business systems is mainly achieved through AI-Agents for data analysis. By leveraging the powerful natural language understanding capabilities of large language models, the business questions raised by users are parsed into structured intents and key entities. The parsed intents are then converted into executable Structured Query Language (SQL) statements, which are submitted to the database for execution. The returned results are then presented to the user in the form of tables or charts.

[0004] However, general-purpose AI agents, designed to serve a wide range of business scenarios, often use generic training data, targeting numerous business lines and failing to focus on a specific business or domain. Consequently, when faced with specific business terminology, metric definitions, and complex business logic, these agents struggle to accurately capture the user's true analytical intent. Their semantic understanding lacks breadth and depth, and accurately mapping natural language into syntactically correct and logically sound SQL is an extremely challenging task. Existing generated SQL statements frequently contain syntactic or logical errors, leading to query failures or incorrect results. The accuracy and robustness of SQL conversion are also suboptimal. Furthermore, these AI agents exhibit weak analytical and insightful capabilities, lacking the ability to proactively uncover underlying relationships within the data. They are unable to perform multi-dimensional drill-down, comparison, and attribution analysis based on query results, making it difficult to provide commercially valuable in-depth insights. Summary of the Invention

[0005] This application provides an insight analysis method and electronic device based on an observation platform to address the shortcomings of existing technologies, such as poor semantic understanding capabilities of general intelligent agents and low conversion accuracy of SQL statements. It can improve the semantic understanding capabilities of intelligent agents, effectively avoid common logical divergences and reasoning loopholes in general models, as well as data ambiguity caused by human misunderstanding or unclear indicator definitions. Thus, a task that originally required generating extremely long and complex SQL statements is broken down into multiple steps that only require calling specific tools, greatly reducing the probability of SQL statement generation errors, realizing fully automated execution, and improving the efficiency of data analysis.

[0006] This application provides an insight analysis method based on an observation platform, comprising: identifying the intent of a received insight analysis request; generating an insight analysis execution path containing multi-step analysis instructions based on the identified task intent and a pre-built structured analysis process configuration library; wherein the pre-built structured analysis process configuration library is constructed based on insight analysis processes corresponding to different task intents; extracting intent keywords from the insight analysis request based on the identified task intent, and performing semantic retrieval in a long-term memory vector library storing business terms and indicator definitions to obtain context information; and triggering the invocation of target tools matching the current step step by step according to the multi-step analysis instructions in the insight analysis execution path based on the context information, and using the execution results of the corresponding tools as input for the corresponding subsequent steps to obtain the insight analysis results.

[0007] According to the insight analysis method based on the observation platform provided in this application, based on context information, the method triggers the invocation of the target tool matching the current step step by step according to the multi-step analysis instructions in the insight analysis execution path. The method includes: for each step analysis instruction in the insight analysis execution path, identifying the instruction intent of the corresponding step analysis instruction, and based on the instruction intent and context information, searching a preset tool library to determine the target tool for the corresponding step analysis instruction; and extracting parameter entities from the insight analysis request and context information according to the selected target tool, generating structured parameters that conform to the parameter format of the corresponding target tool, and configuring the corresponding target tool.

[0008] According to the insight analysis method based on an observation platform provided in this application, a target tool is selected from a candidate tool set corresponding to a step analysis instruction based on context information and the insight analysis execution path. This includes: selecting a target tool from the candidate tool set corresponding to the step analysis instruction based on context information and the insight analysis execution path, combined with a preset selection strategy; wherein the preset selection strategy is used to: determine the degree of fit between each candidate tool in the candidate tool set and the corresponding step analysis instruction, obtaining the functional matching degree of each candidate tool; determine whether the required parameters of the corresponding candidate tool can be completely extracted from the context information based on each candidate tool in the candidate tool set, the insight analysis request, and the context information, obtaining the parameter compatibility of each candidate tool; obtain the historical execution success rate and / or estimated execution cost of each candidate tool, obtaining the reliability evaluation result of each candidate tool; and comprehensively evaluate each candidate tool based on the functional matching degree, the parameter compatibility, and the reliability evaluation result of each candidate tool to determine the corresponding target tool.

[0009] According to the insight analysis method based on an observation platform provided in this application, parameter entities are extracted from the insight analysis request and context information based on the selected target tool, and structured parameters conforming to the parameter format of the corresponding target tool are generated. The method includes: extracting parameter entities from the insight analysis request and context information based on the selected target tool; searching for matching nodes in a pre-built business knowledge graph based on the extracted parameter entities to obtain candidate nodes; wherein the business knowledge graph is constructed based on entities of different business data and the relationships between these entities, and each node of the business knowledge graph stores corresponding entity transformation rules configured based on different tool parameter formats; performing context disambiguation based on the candidate nodes and context information to obtain target nodes, and linking the corresponding parameter entities to the target nodes of the business knowledge graph; and searching for the transformation rules of the corresponding target tool in the business knowledge graph based on the target nodes, and performing transformation on the corresponding parameter entities according to the found transformation rules to obtain structured parameters conforming to the parameter format of the corresponding target tool.

[0010] According to the insight analysis method based on the observation platform provided in this application, after extracting intent keywords from the insight analysis request based on the identified intent and performing semantic retrieval in a long-term memory vector library storing business terms and indicator definitions to obtain context information, the method includes: matching and retrieving the identified intent and insight analysis request with a short-term memory library recording the current conversation reasoning chain to obtain historical interaction information corresponding to the dialogue turn; and fusing the context information and the historical interaction information corresponding to the dialogue turn.

[0011] According to the insight analysis method based on the observation platform provided in this application, information fusion is performed on contextual information and historical interaction information corresponding to dialogue turns, including: treating contextual information as static knowledge, treating historical interaction information corresponding to dialogue turns as dynamic dialogue context, and determining the attention weights of static knowledge and dynamic dialogue context in combination with the intent identified in the insight analysis request; and performing weighted aggregation based on static knowledge and dynamic dialogue context, combined with the corresponding attention weights, to obtain contextual information that fuses static knowledge and dynamic dialogue context.

[0012] According to the insight analysis method based on an observation platform provided in this application, an insight analysis execution path containing multi-step analysis instructions is generated based on the identified task intent and in conjunction with a preset structured analysis process configuration library. The method includes: determining the task type based on the identified intent and obtaining the corresponding insight analysis process configuration file in conjunction with the preset structured analysis process configuration library; parsing the insight analysis process configuration file to obtain a process topology structure containing multiple analysis step nodes, and initializing a dynamic execution context for recording the execution status and intermediate results of each node based on the business parameters corresponding to the insight analysis request; and dynamically adjusting the process topology structure according to the dynamic execution context and a preset optimization strategy to generate the insight analysis execution path. The dynamic adjustment includes pruning nodes that do not meet the execution conditions, attaching new analysis nodes, or selecting the optimal path from multiple executable branches.

[0013] According to the insight analysis method based on the observation platform provided in this application, the process topology includes a directed acyclic graph (DAG). Each node in the DAG encapsulates the corresponding step analysis instruction, input parameter specifications, and decision conditions for jumping to the next node. Based on the business parameters corresponding to the insight analysis request, a dynamic execution context for recording the execution status and intermediate results of each node is initialized. This includes: a general business parameter entity extracted from the insight analysis request, and parameter binding of the extracted general business parameter entity to the starting node of the DAG to initialize the dynamic execution context for recording the execution status and intermediate results of each node.

[0014] According to the insight analysis method based on an observation platform provided in this application, the intent recognition of received insight analysis requests includes: loading a preset agent configuration file and converting the content of the preset agent configuration file into a structured object; wherein the preset agent configuration file is configured in advance based on the role and responsibility of the agent; extracting key features from the structured object and filling the extracted key features into a preset system prompt template to generate a system prompt; and performing intent recognition on the received insight analysis request based on the system prompt.

[0015] This application also provides an insight analysis device based on an observation platform, comprising: a path generation module, which identifies the intent of a received insight analysis request and generates an insight analysis execution path containing multi-step analysis instructions based on the identified task intent and a preset structured analysis process configuration library; wherein the preset structured analysis process configuration library is constructed in advance based on insight analysis processes corresponding to different task intents; a context acquisition module, which extracts intent keywords from the insight analysis request based on the identified task intent and performs semantic retrieval in a long-term memory vector library storing business terms and indicator definitions to obtain context information; and an insight analysis module, which, based on the context information, triggers the invocation of target tools matching the current step step by step according to the multi-step analysis instructions in the insight analysis execution path, and uses the execution results of the corresponding tools as input for the corresponding subsequent steps to obtain the insight analysis results.

[0016] This application also provides an electronic device including a memory and a processor, wherein the memory stores a computer program and the processor is configured to execute the observation platform-based insight analysis method as described above through the computer program.

[0017] This application also provides a computer-readable storage medium comprising a stored program, wherein the program, when executed, implements the insight analysis method based on the observation platform as described above.

[0018] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the insight analysis method based on the observation platform as described above.

[0019] The insight analysis method and electronic device based on the observation platform provided in this application improve the semantic understanding ability of the intelligent agent by identifying the intent of the received insight analysis request and extracting intent keywords based on the identified task intent. This enables the agent to accurately understand the user's natural language request, lowers the technical threshold for users to use data analysis, and generates an execution path based on a preset structured process configuration library. This ensures that the analysis logic for specific tasks is standardized and verified, avoiding logical divergence and reasoning loopholes common in general models. Furthermore, it utilizes a long-term memory vector library to retrieve business terms and indicator definitions, transforming the user's natural language request into contextual information with clear business definitions. This ensures that the analysis process is based on a unified business standard within the enterprise, effectively avoiding data ambiguity caused by human misunderstanding or unclear indicator definitions. Based on the retrieved contextual information, the target toolchain is triggered according to the insight analysis execution path. This breaks down a task that originally required generating extremely long and complex SQL statements into multiple steps, with each step only requiring the invocation of specific tools. This significantly reduces the probability of SQL statement generation errors, achieves fully automated execution, and improves the efficiency of data analysis. Attached Figure Description

[0020] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0021] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the hardware environment for an insight analysis method based on an observation platform according to an embodiment of this application; Figure 2 This is a flowchart illustrating the insight analysis method based on the observation platform provided in this application; Figure 3 This is a schematic diagram of the structure of the observation platform-based insight analysis device provided in this application; Figure 4 This is a schematic diagram of the electronic device provided in this application. Detailed Implementation

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

[0024] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0025] According to one aspect of the embodiments of this application, an insight analysis method based on an observation platform is provided. This insight analysis method based on an observation platform is widely applicable to whole-house intelligent digital control application scenarios such as smart homes, smart home ecosystems, and intelligence house ecosystems. Optionally, in this embodiment, the above-mentioned insight analysis method based on an observation platform can be applied to, for example... Figure 1 The hardware environment shown consists of terminal device 102 and server 104. For example... Figure 1 As shown, server 104 is connected to terminal device 102 via a network and can be used to provide services (such as application services) to the terminal or clients installed on the terminal. A database can be set up on the server or independently of the server to provide data storage services for server 104. Cloud computing and / or edge computing services can be configured on the server or independently of the server to provide data processing services for server 104.

[0026] The aforementioned network may include, but is not limited to, at least one of the following: wired network, wireless network. The aforementioned wired network may include, but is not limited to, at least one of the following: wide area network, metropolitan area network, local area network. The aforementioned wireless network may include, but is not limited to, at least one of the following: Wi-Fi (Wireless Fidelity), Bluetooth. The terminal device 102 may not be limited to PC, mobile phone, tablet computer, smart air conditioner, smart range hood, smart refrigerator, smart oven, smart stove, smart washing machine, smart water heater, smart washing equipment, smart dishwasher, smart projector, smart TV, smart clothes rack, smart curtains, smart audio-visual equipment, smart socket, smart speaker, smart speaker box, smart fresh air equipment, smart kitchen and bathroom equipment, smart bathroom equipment, smart robot vacuum cleaner, smart window cleaning robot, smart mopping robot, smart air purifier, smart steam oven, smart microwave oven, smart water heater, smart air purifier, smart water dispenser, smart door lock, etc.

[0027] Figure 2 This is one of the flowcharts illustrating the insight analysis method based on an observation platform provided by the present invention, such as... Figure 2 As shown, the method includes the following: S21, the received insight analysis request is subjected to intent recognition, and based on the recognized task intent, combined with the preset structured analysis process configuration library, an insight analysis execution path containing multi-step analysis instructions is generated; wherein, the preset structured analysis process configuration library is built in advance based on the insight analysis process corresponding to different task intents; S22, based on the identified task intent, extract intent keywords from the insight analysis request, and perform semantic retrieval in a long-term memory vector library that stores business terms and indicator definitions to obtain contextual information; S23. Based on the context information, according to the multi-step analysis instructions in the insight analysis execution path, the target tool matching the current step is called step by step, and the execution result of the corresponding tool is used as the input for the corresponding subsequent step to obtain the insight analysis result.

[0028] It should be noted that the step number "S2N" in this specification does not represent the order of the insight analysis methods based on the observation platform. The insight analysis method based on the observation platform of the present invention is described in detail below.

[0029] Step S21: The received insight analysis request is subjected to intent recognition, and an insight analysis execution path containing multi-step analysis instructions is generated based on the recognized task intent and in conjunction with the preset structured analysis process configuration library; wherein, the preset structured analysis process configuration library is constructed in advance based on the insight analysis process corresponding to different task intents.

[0030] In this embodiment, the intent recognition of the received insight analysis request includes: loading a preset agent configuration file and converting the contents of the preset agent configuration file into a structured object; wherein the preset agent configuration file is configured in advance based on the role and responsibility of the agent; extracting key features from the structured object and filling the extracted key features into a preset system prompt template to generate a system prompt; and performing intent recognition on the received insight analysis request based on the system prompt.

[0031] It should be noted that by loading and converting the pre-set intelligent agent configuration file, unstructured or semi-structured configuration text is parsed into objects that can be directly manipulated by the computer. This optimizes the maintainability and hot update capability of the intelligent agent system, reduces the human cost of system iteration and adjustment, and extracts key features from the structured objects to populate the pre-set system prompt template. This results in a customized system prompt that integrates the static identity and dynamic instructions of the intelligent agent. When handling fuzzy requests, it can prioritize matching words within its responsibility scope, thereby optimizing the contextual understanding depth of the intent recognition module and significantly improving the accuracy of intent recognition. Based on the system prompts, intent recognition can be performed under specific business context constraints, improving the accuracy of intent recognition and optimizing the robustness of the entire insight analysis process in complex business scenarios.

[0032] It should be added that the key features include name features, role features, professional features, and responsibility features. Name features are used to represent the identity of the intelligent agent, role features are used to represent the social or functional identity played by the intelligent agent in the target scenario or dialogue, professional features are used to represent the knowledge domain and skill set used by the intelligent agent, and responsibility features are used to represent the specific responsibilities and task list that the intelligent agent needs to perform.

[0033] In addition, based on the system prompts, intent recognition is performed on the received insight analysis requests, including: inputting the system prompts and the received insight analysis requests into the large language model to perform intent recognition on the insight analysis requests based on the system prompts.

[0034] Furthermore, after performing intent recognition on the received insight analysis request according to the system prompts, the process includes: when the confidence level of the recognized intent is less than a preset confidence threshold, or when the recognized intent has multiple ambiguous candidate branches, generating a clarifying question based on a business terminology dictionary and returning it to the user to obtain further input, until the intent is uniquely determined; wherein, the business terminology dictionary includes standardized names of indicators within the observation platform, synonym mappings, and terminology explanations.

[0035] Specifically, clarifying questions are generated based on a business terminology dictionary, including: when the intent is determined to be a polysemous entity, the standardized names and terminology explanations of all candidate indicators are extracted from the business terminology dictionary, and clarifying questions are generated in the form of options to clearly present the options to the user; when the intent cannot be directly matched in the business terminology dictionary, but related terms can be obtained through fuzzy matching or association queries, the intent is determined to be a fuzzy entity, and based on the terminology explanations, a preset number of indicators most similar to the intent are found, and clarifying questions are generated in a guided manner to help the user locate the problem; when the intent is clear but the core entity of the insight analysis request is missing, the indicators that the user may be interested in are predicted from contextual information or historical records, and clarifying questions are generated using the standardized names in the business terminology dictionary, or an open-ended clarifying question is generated using the standardized names in the business terminology dictionary to guide the user to use standard terms.

[0036] Furthermore, in this embodiment, based on the identified task intent and in conjunction with a preset structured analysis process configuration library, an insight analysis execution path containing multi-step analysis instructions is generated. This includes: determining the task type based on the identified intent and obtaining the corresponding insight analysis process configuration file based on the preset structured analysis process configuration library; parsing the insight analysis process configuration file to obtain a process topology structure containing multiple analysis step nodes, and initializing a dynamic execution context for recording the execution status and intermediate results of each node based on the business parameters corresponding to the insight analysis request; and dynamically adjusting the process topology structure based on the dynamic execution context and a preset optimization strategy to generate the insight analysis execution path. The dynamic adjustment includes pruning nodes that do not meet the execution conditions, attaching new analysis nodes, or selecting the optimal path from multiple executable branches.

[0037] Specifically, the preset optimization strategy is used to: prune path nodes and their subsequent branches when it is determined that the general business parameter entity in the dynamic execution context does not meet the execution preconditions of the corresponding path node; load and attach new analysis nodes to the current execution path when it is determined that the intermediate results in the dynamic execution context meet the preset exception patterns or heuristic rules; and select the path branch with the lowest overall cost or the highest expected benefit based on the node computation cost and data query time or estimated value of each path branch when it is determined that there are multiple parallel path branches.

[0038] In addition, the process topology includes a directed acyclic graph (DAG). Each node in the DAG encapsulates the corresponding step analysis instructions, input parameter specifications, and decision conditions for jumping to the next node. Accordingly, based on the business parameters corresponding to the insight analysis request, a dynamic execution context for recording the execution status and intermediate results of each node is initialized. This includes: a general business parameter entity extracted from the insight analysis request, and parameter binding of the extracted general business parameter entity to the starting node of the DAG to initialize the dynamic execution context for recording the execution status and intermediate results of each node.

[0039] It should be noted that by parsing the insight analysis process configuration file, a structured and visualized analysis logic topology is obtained, transforming the originally linear and difficult-to-maintain analysis process into a graph model with conditional judgment and branching capabilities. This optimizes the controllability and logical rigor of the analysis process in complex business scenarios. Furthermore, by binding common business parameter entities, an initial running instance is obtained that connects specific business problems with the general analysis process. Instantiation and initialization are achieved through dynamic execution context to realize full-process state tracking, thus completely decoupling business logic from technology and optimizing the versatility and reusability of the analysis system.

[0040] It should be added that when executing any node in the personalized analysis execution path, the node is configured to: dynamically match and select one or more tools from the preset tool registry based on the business parameter entities required by the current node, and use the conversion rules embedded in the node corresponding to the tools to convert the business parameter entities into dedicated structured parameters that conform to the parameter format of the selected target tool in order to complete the task of the node. For details, please refer to the following text, and no further limitations are made here.

[0041] Step S22: Based on the identified task intent, extract intent keywords from the insight analysis request and perform semantic retrieval in a long-term memory vector library that stores business terms and indicator definitions to obtain contextual information.

[0042] In this embodiment, based on the identified intent, intent keywords are extracted from the insight analysis request, and semantic retrieval is performed in a long-term memory vector library that stores business terms and indicator definitions to obtain contextual information. This includes: extracting intent keywords from the insight analysis request based on the identified intent and converting them into high-dimensional vectors; using the high-dimensional vectors, performing semantic similarity search in the long-term memory vector library, and selecting the target number of knowledge items from the query results returned by the long-term memory vector library in descending order of similarity to obtain contextual information.

[0043] Furthermore, after extracting intent keywords from the insight analysis request based on the identified intent and performing semantic retrieval in a long-term memory vector library storing business terms and indicator definitions to obtain context information, the process includes: matching and retrieving the identified intent and insight analysis request with a short-term memory library recording the current conversation reasoning chain to obtain historical interaction information corresponding to the dialogue turn; and fusing the context information and the historical interaction information corresponding to the dialogue turn.

[0044] Furthermore, the information fusion of contextual information and historical interaction information corresponding to the dialogue rounds includes: treating the contextual information as static knowledge, treating the historical interaction information corresponding to the dialogue rounds as dynamic dialogue context, and combining the intention identified by the insight analysis request to determine the attention weights of the static knowledge and dynamic dialogue context; and performing weighted aggregation based on the static knowledge and dynamic dialogue context, combined with the corresponding attention weights, to obtain contextual information that fuses the static knowledge and dynamic dialogue context.

[0045] It should be noted that by utilizing the high-dimensional vector of the identified intent, semantic similarity search is performed in the long-term memory vector library to effectively bridge the gap between user expression and professional terminology, ensuring that the extracted knowledge is the business meaning that the user truly wants. Furthermore, the identified intent and insight analysis request are matched and retrieved with the short-term memory library that records the current conversation reasoning chain, obtaining a dual information input source from the Russian that includes static domain knowledge and dynamic conversation state. Long-term memory ensures the professionalism and authority of the analysis basis, while short-term memory ensures the logical coherence of the dialogue. Combining the two significantly reduces the difficulty of model reasoning, enabling it to generate more accurate analysis instructions and insight results that better meet user expectations, thereby improving the execution effect and accuracy of subsequent insight analysis tasks.

[0046] Step S23: Based on the context information, trigger the call of the target tool matching the current step step by step according to the multi-step analysis instructions in the insight analysis execution path, and use the execution result of the corresponding tool as the input of the corresponding subsequent step to obtain the insight analysis result.

[0047] In this embodiment, based on context information, the target tool matching the current step is triggered step by step according to the multi-step analysis instructions in the insight analysis execution path. This includes: identifying the instruction intent of the corresponding step analysis instruction for each step analysis instruction in the insight analysis execution path, and searching a preset tool library based on the instruction intent and context information to determine the target tool for the corresponding step analysis instruction; extracting parameter entities from the insight analysis request and context information according to the selected target tool, generating structured parameters that conform to the parameter format of the corresponding target tool, and configuring the corresponding target tool.

[0048] Furthermore, based on the instruction intent and context information, a preset tool library is searched to determine the target tool for the corresponding step analysis instruction. This includes: determining the semantic similarity between the instruction intent and the functional description vectors of each tool in the preset tool library, and filtering tools with similarity higher than a first preset threshold to obtain a candidate tool set corresponding to the step analysis instruction; and selecting the target tool from the candidate tool set corresponding to the step analysis instruction based on context information and insight analysis execution path.

[0049] It should be noted that by screening candidate tools in the above manner, a candidate tool set based on deep semantic understanding rather than simple keyword matching is obtained. This effectively avoids tool matching failures caused by differences in the wording of user or system instructions, improves the accuracy and recall rate of tool retrieval, and further selects target tools from the candidate tool set corresponding to the step analysis instructions based on contextual information and insight analysis of the execution path. This ensures that the selected target tools meet the functional requirements of the current step while being compatible with the current session state and subsequent process planning, ensuring the rationality of tool selection and execution consistency. Parameters are then extracted and formatted for the selected target tools to eliminate the grammatical gap between business intent and technical implementation. This automatically transforms natural language requests and general parameters at the business level into execution instances of API call parameters at the technical level, reducing call failures caused by incorrect parameter formatting and significantly improving the efficiency of the agent in executing tasks.

[0050] Specifically, based on contextual information and the insight analysis execution path, a target tool is selected from the candidate tool set corresponding to the step analysis instruction. This includes: selecting a target tool from the candidate tool set corresponding to the step analysis instruction based on contextual information and the insight analysis execution path, combined with a preset selection strategy; wherein, the preset selection strategy is used to: determine the degree of fit between each candidate tool in the candidate tool set and the corresponding step analysis instruction, thereby obtaining the functional matching degree of each candidate tool; determine whether the required parameters of the corresponding candidate tool can be completely extracted from the context information based on each candidate tool in the candidate tool set, the insight analysis request, and the context information, thereby obtaining the parameter compatibility of each candidate tool; obtain the historical execution success rate and / or estimated execution cost of each candidate tool, thereby obtaining the reliability assessment result of each candidate tool; and comprehensively evaluate each candidate tool based on the functional matching degree, parameter compatibility, and reliability assessment results of each candidate tool to determine the corresponding target tool.

[0051] In addition, based on the selected target tool, parameter entities are extracted from the insight analysis request and context information to generate structured parameters that conform to the parameter format of the corresponding target tool. This includes: extracting parameter entities from the insight analysis request and context information based on the selected target tool; searching for matching nodes in a pre-built business knowledge graph based on the extracted parameter entities to obtain candidate nodes; wherein, the business knowledge graph is constructed based on entities of different business data and the relationships between entities, and each node of the business knowledge graph stores corresponding entity transformation rules configured based on different tool parameter formats; based on the candidate nodes, context disambiguation is performed in conjunction with context information to obtain target nodes, and the corresponding parameter entities are linked to the target nodes of the business knowledge graph; based on the target nodes, the transformation rules of the corresponding target tool in the business knowledge graph are found, and the corresponding parameter entities are transformed according to the found transformation rules to obtain structured parameters that conform to the parameter format of the corresponding target tool.

[0052] Furthermore, based on the candidate nodes, context disambiguation is performed in conjunction with context information to obtain the target node, including: determining the semantic similarity between the context information and each candidate node based on the context information and each candidate node; determining whether the data type of the candidate node is compatible with the parameter type of the target tool to obtain the type compatibility of each candidate node; and determining the target node based on the previously obtained historical usage frequency of the node and / or the optimal path of the graph, combined with the semantic similarity between the context information and each candidate node and the type compatibility of each candidate node.

[0053] In an optional embodiment, after linking the corresponding parameter entity to the target node of the business knowledge graph, attribute and relationship disambiguation can also be performed. Specifically, based on the target node, the relationship nodes of the target node in the business knowledge graph are traversed to determine the scope and level of the target node in the business knowledge graph, and the relationship of the corresponding target node is obtained.

[0054] It should be noted that by traversing the relational nodes of the target node in the business knowledge graph, the ambiguity caused by homonyms or polysemy is effectively resolved. This ensures that subsequent analysis is based on the data granularity and business scope that users truly care about, optimizing the depth and accuracy of the agent's understanding of business concepts. This determines the scope and level of the target node in the business knowledge graph. By clarifying the relationships between entities, the agent can not only query single indicators but also perform correlation analysis, thereby uncovering the business logic chain behind the data, providing more valuable comprehensive insights, achieving disambiguation of relationships and attributes, greatly reducing the interference of invalid information on the model, and improving the accuracy of subsequent tool selection and parameter generation.

[0055] In addition, before searching for matching nodes in the pre-built business knowledge graph based on the extracted parameter entities to obtain candidate nodes, the process includes: acquiring business data; wherein, business data includes at least one of the following: business database, application programming interface (API) documentation, monitoring configuration files, preset code libraries, data dictionaries or indicator dictionaries, technical documents, IM chat logs or work order systems, and expert knowledge; extracting entities, relationships, and attributes from the business data; merging entities from different sources that point to the same concept; determining the corresponding indicator definition based on preset data source priority or artificial intelligence when different sources have different definitions for the same indicator; inferring implicit knowledge from existing knowledge based on the extracted entities, relationships, and attributes, and performing knowledge completion on existing knowledge; and storing the obtained knowledge according to the pre-selected storage type to form a queryable business knowledge graph.

[0056] It should be added that the business database schema includes MySQL / PostgreSQL CREATE TABLE statements, field comments, and field types; API documentation includes API development and documentation management tools or API descriptions (Swagger / OpenAPI) specifications, defining interface paths, parameters, and return value structures; monitoring configuration files include configuration files for the open-source system monitoring and alerting toolkit Prometheus, JSON definitions for Grafana dashboard configuration files, log collection rules, etc., which directly contain valuable information such as metric names, tags, and calculation expressions; pre-built code repositories, such as parsing Python and Java code, especially SQL query statements and ORM model definitions, which can deduce the relationships between data tables; data dictionaries include internally maintained Excel or Wiki pages, clearly defining the meaning, calculation formula, and responsible person for each business metric; technical documentation includes system architecture diagrams, business process descriptions, and troubleshooting manuals on platforms such as Confluence and Notion; IM chat logs / work order systems include discussion records of engineers solving problems, containing a large amount of colloquial descriptions of problems and corresponding solutions; expert knowledge includes interviews with domain experts and historical analysis reports, etc.

[0057] In addition, entity extraction includes: extracting database names, table names, and field names (such as device_status, online_count) as entities from technical metadata; and extracting business terms as entities from business documents using NLP techniques. Relationship extraction includes: extracting foreign key relationships between tables from the business database schema or architecture; parsing SQL JOIN statements to extract table relationships from the code; and using relationship extraction models to identify relationships between entities in documents, such as the relationship "the number of online devices is a core indicator of device health." Attribute extraction includes: extracting the business meaning of fields from field comments (the attribute of online_count is {description: "total number of online devices", unit: "devices"}); and extracting attributes such as parameter type and whether they are required from API documentation.

[0058] It is worth noting that after each invocation of the target tool and return of the corresponding tool execution result, it is stored in a short-term memory to record the session context.

[0059] In summary, this invention improves the semantic understanding of intelligent agents by identifying the intent of received insight analysis requests and extracting intent keywords based on the identified task intent. This enables the agents to accurately understand users' natural language requests, lowering the technical threshold for users to use data analysis. Furthermore, it generates execution paths based on a pre-defined structured process configuration library, ensuring that the analysis logic for specific tasks is standardized and validated, avoiding common logical divergences and reasoning loopholes in general models. It also utilizes a long-term memory vector library to retrieve business terms and indicator definitions, transforming users' natural language requests into contextual information with clear business definitions. This ensures that the analysis process is based on a unified business standard within the enterprise, effectively avoiding data ambiguity caused by human misunderstanding or unclear indicator definitions. Based on the retrieved contextual information, the target toolchain is triggered according to the insight analysis execution path. This breaks down a task that originally required generating extremely long and complex SQL statements into multiple steps, each requiring only the invocation of specific tools, significantly reducing the probability of SQL statement generation errors and achieving fully automated execution, thus improving the efficiency of data analysis.

[0060] The observation platform-based insight analysis device provided in this application is described below. The observation platform-based insight analysis device described below and the observation platform-based insight analysis method described above can be referred to in correspondence.

[0061] Figure 3 A schematic diagram of an insight analysis device based on an observation platform is shown. The device includes: The path generation module 31 performs intent recognition on the received insight analysis request, and generates an insight analysis execution path containing multi-step analysis instructions based on the recognized task intent and a preset structured analysis process configuration library; wherein, the preset structured analysis process configuration library is built in advance based on the insight analysis process corresponding to different task intents; The context acquisition module 32 extracts intent keywords from the insight analysis request based on the identified task intent, and performs semantic retrieval in a long-term memory vector library that stores business terms and indicator definitions to obtain context information. The insight analysis module 33, based on context information, triggers the invocation of the target tool matching the current step step by step according to the multi-step analysis instructions in the insight analysis execution path, and uses the execution result of the corresponding tool as the input for the corresponding subsequent step to obtain the insight analysis result.

[0062] In this embodiment, the path generation module 31 includes: a parsing unit, which loads a preset agent configuration file and converts the contents of the preset agent configuration file into a structured object; wherein the preset agent configuration file is configured in advance based on the role and responsibility of the agent; a prompt generation unit, which extracts key features from the structured object and fills the extracted key features into a preset system prompt template to generate a system prompt; and an intent recognition unit, which performs intent recognition on the received insight analysis request according to the system prompt.

[0063] It should be added that the intent recognition unit is used to: input system prompts and received insight analysis requests into the large language model to perform intent recognition on the insight analysis requests based on the system prompts.

[0064] Furthermore, the path generation module 31 also includes: an intent acquisition unit, which, after identifying the intent of the received insight analysis request according to the system prompts, determines that the confidence level of the identified intent is less than a preset confidence threshold, or that the identified intent has multiple ambiguous candidate branches, generates a clarifying question based on a business terminology dictionary, and returns it to the user to obtain further input, until the intent is uniquely determined; wherein, the business terminology dictionary module includes standardized names of indicators within the observation platform, synonym mappings, and terminology explanations.

[0065] Specifically, the intent acquisition unit is used to: when the intent is determined to be a polysemous entity, extract standardized names and terminology explanations of all candidate indicators from the business terminology dictionary, and generate clarifying questions in the form of options to clearly present the options to the user; when the intent cannot be directly matched in the business terminology dictionary, but related terms can be obtained through fuzzy matching or association queries, determine that the intent is a fuzzy entity, find a preset number of indicators most similar to the intent based on the terminology explanation, and generate clarifying questions in a guided manner to help the user locate the problem; when the intent is clear but the core entity of the insight analysis request is missing, predict the indicators that the user may care about from contextual information or historical records, and generate clarifying questions using standardized names in the business terminology dictionary, or generate an open-ended clarifying question using standardized names in the business terminology dictionary to guide the user to use standard terms.

[0066] Furthermore, in this embodiment, the path generation module 31 further includes: a file acquisition unit, which determines the task type based on the identified intent and acquires the corresponding insight analysis process configuration file in conjunction with a preset structured analysis process configuration library; an initialization configuration unit, which parses the insight analysis process configuration file to obtain a process topology structure containing multiple analysis step nodes, and initializes a dynamic execution context for recording the execution status and intermediate results of each node according to the business parameters corresponding to the insight analysis request; and a path generation unit, which dynamically adjusts the process topology structure according to the dynamic execution context and a preset optimization strategy to generate an insight analysis execution path; wherein, the dynamic adjustment includes pruning nodes that do not meet the execution conditions, attaching new analysis nodes, or selecting the optimal path from multiple executable branches.

[0067] Specifically, the preset optimization strategy is used to: prune path nodes and their subsequent branches when it is determined that the general business parameter entity in the dynamic execution context does not meet the execution preconditions of the corresponding path node; load and attach new analysis nodes to the current execution path when it is determined that the intermediate results in the dynamic execution context meet the preset exception patterns or heuristic rules; and select the path branch with the lowest overall cost or the highest expected benefit based on the node computation cost and data query time or estimated value of each path branch when it is determined that there are multiple parallel path branches.

[0068] In addition, the process topology includes a directed acyclic graph (DAG), where each node encapsulates the corresponding step analysis instructions, input parameter specifications, and decision conditions for jumping to the next node. Correspondingly, an initialization configuration unit is used to: extract general business parameter entities from the insight analysis request and bind these entities to the starting node of the DAG to initialize a dynamic execution context for recording the execution status and intermediate results of each node.

[0069] The context acquisition module 32 includes: a vector conversion unit, which extracts intent keywords from the insight analysis request and converts them into high-dimensional vectors based on the identified intent; and a first retrieval unit, which uses the high-dimensional vectors to perform semantic similarity searches in the long-term memory vector library, and selects the target number of knowledge items from the query results returned by the long-term memory vector library in descending order of similarity to obtain context information.

[0070] Furthermore, the context acquisition module 32 also includes: a second retrieval unit, which, after extracting intent keywords from the insight analysis request based on the identified intent and performing semantic retrieval in a long-term memory vector library storing business terms and indicator definitions to obtain context information, matches and retrieves the identified intent and insight analysis request with a short-term memory library recording the current conversation reasoning chain to obtain historical interaction information corresponding to the dialogue turn; and an information fusion unit, which fuses the context information and the historical interaction information corresponding to the dialogue turn.

[0071] Furthermore, the information fusion unit includes: a weight determination subunit, which uses contextual information as static knowledge, uses historical interaction information corresponding to dialogue turns as dynamic dialogue context, and combines the intent identified by the insight analysis request to determine the attention weights of static knowledge and dynamic dialogue context; and an information fusion subunit, which performs weighted aggregation based on static knowledge and dynamic dialogue context, combined with the corresponding attention weights, to obtain contextual information that integrates static knowledge and dynamic dialogue context.

[0072] The insight analysis module 33 includes: a tool determination unit, which identifies the instruction intent of the corresponding step analysis instruction for each step of the insight analysis execution path, and searches a preset tool library based on the instruction intent and context information to determine the target tool for the corresponding step analysis instruction; and a parameter configuration unit, which extracts parameter entities from the insight analysis request and context information according to the selected target tool, generates structured parameters that conform to the parameter format of the corresponding target tool, and configures the corresponding target tool.

[0073] Furthermore, the tool determination unit includes: a candidate tool screening subunit, which determines the semantic similarity between the instruction intent and the functional description vectors of each tool in the preset tool library, and filters tools with similarity higher than a first preset threshold to obtain a candidate tool set corresponding to the step analysis instruction; and a target tool determination subunit, which selects a target tool from the candidate tool set corresponding to the step analysis instruction based on context information and insight analysis execution path.

[0074] Specifically, the target tool determination unit is used to: select a target tool from the candidate tool set corresponding to the step analysis instruction based on context information and the insight analysis execution path, combined with a preset selection strategy; wherein, the preset selection strategy is used to: determine the degree of fit between each candidate tool in the candidate tool set and the corresponding step analysis instruction, thereby obtaining the functional matching degree of each candidate tool; determine whether the required parameters of the corresponding candidate tool can be completely extracted from the context information based on each candidate tool in the candidate tool set, the insight analysis request, and the context information, thereby obtaining the parameter compatibility of each candidate tool; obtain the historical execution success rate and / or estimated execution cost of each candidate tool, thereby obtaining the reliability evaluation result of each candidate tool; and comprehensively evaluate each candidate tool based on the functional matching degree, the parameter compatibility, and the reliability evaluation result of each candidate tool, thereby determining the corresponding target tool.

[0075] Additionally, the parameter generation unit includes: an entity extraction subunit, which extracts parameter entities from the insight analysis request and context information based on the selected target tool; a node selection subunit, which searches for matching nodes in a pre-built business knowledge graph based on the extracted parameter entities to obtain candidate nodes; wherein, the business knowledge graph is constructed in advance based on entities of different business data and the relationships between entities, and each node of the business knowledge graph stores corresponding entity transformation rules configured in advance based on different tool parameter formats; a disambiguation linking subunit, which performs context disambiguation based on candidate nodes and context information to obtain target nodes, and links the corresponding parameter entities to the target nodes of the business knowledge graph; and a parameter generation subunit, which searches for the transformation rules of the corresponding target tool in the business knowledge graph based on the target nodes, and performs transformation on the corresponding parameter entities according to the found transformation rules to obtain structured parameters that conform to the parameter format of the corresponding target tool.

[0076] Furthermore, the disambiguation linking subunit is used to: determine the semantic similarity between the context information and each candidate node based on the context information and each candidate node; determine whether the data type of the candidate node is compatible with the parameter type of the target tool to obtain the type compatibility of each candidate node; and determine the target node based on the previously obtained historical usage frequency of the node and / or the optimal path of the graph, combined with the semantic similarity between the context information and each candidate node and the type compatibility of each candidate node.

[0077] In an optional embodiment, the parameter generation unit further includes an attribute relationship disambiguation subunit. After linking the corresponding parameter entity to the target node of the business knowledge graph, it can also perform attribute and relationship disambiguation. Specifically, it is used to: based on the target node, traverse the relationship nodes of the target node in the business knowledge graph, determine the scope and level of the target node in the business knowledge graph, and obtain the relationship of the corresponding target node.

[0078] In addition, the parameter generation unit further includes: a data acquisition subunit, which acquires business data before searching for matching nodes in a pre-built business knowledge graph based on the extracted parameter entities to obtain candidate nodes; wherein, the business data includes at least one of the following: business database, application programming interface (API) documentation, monitoring configuration files, preset code libraries, data dictionaries or indicator dictionaries, technical documents, IM chat logs or work order systems, and expert knowledge; performs entity extraction, relation extraction, and attribute extraction on the business data; merges entities from different sources that point to the same concept; when different sources have different definitions of the same indicator, determines the corresponding indicator definition according to the preset data source priority or artificial intelligence; infers the implicit knowledge of existing knowledge based on the extracted entities, relations, and attributes, and performs knowledge completion on the existing knowledge; and stores the knowledge obtained above according to the pre-selected storage type to form a queryable business knowledge graph.

[0079] In summary, this embodiment of the invention uses a path generation module to identify the intent of received insight analysis requests and a context acquisition module to extract intent keywords based on the identified task intent. This enhances the semantic understanding capability of the intelligent agent, enabling it to accurately understand the user's natural language requests and lowering the technical threshold for users to use data analysis. The path generation module generates execution paths based on a preset structured process configuration library, ensuring that the analysis logic for specific tasks is standardized and validated, avoiding common logical divergences and reasoning loopholes in general models. The context acquisition module uses a long-term memory vector library to retrieve business terms and indicator definitions, transforming the user's natural language requests into contextual information with clear business definitions. This ensures that the analysis process is based on a unified business standard within the enterprise, effectively avoiding data ambiguity caused by human misunderstanding or unclear indicator definitions. Furthermore, the insight analysis module, based on the retrieved context information and the insight analysis execution path, triggers the target toolchain, thus breaking down a task that originally required generating extremely long and complex SQL statements into multiple steps. Each step only needs to call a specific tool, greatly reducing the probability of SQL statement generation errors and achieving fully automated execution, thereby improving the efficiency of data analysis.

[0080] Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4As shown, the electronic device may include a processor 410, a communications interface 420, a memory 430, and a communication bus 440. 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 an insight analysis method based on an observation platform. This method includes: identifying the intent of a received insight analysis request; generating an insight analysis execution path containing multi-step analysis instructions based on the identified task intent and a pre-built structured analysis process configuration library; wherein the pre-built structured analysis process configuration library is constructed based on insight analysis processes corresponding to different task intents; extracting intent keywords from the insight analysis request based on the identified task intent and performing semantic retrieval in a long-term memory vector library storing business terms and indicator definitions to obtain context information; and triggering the invocation of target tools matching the current step step by step according to the multi-step analysis instructions in the insight analysis execution path based on the context information, and using the execution results of the corresponding tools as input for the corresponding subsequent steps to obtain the insight analysis result.

[0081] 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 this application, in essence, 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 this application. 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.

[0082] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the insight analysis method based on the observation platform provided by the above methods. The method includes: identifying the intent of a received insight analysis request, and generating an insight analysis execution path containing multi-step analysis instructions based on the identified task intent and a preset structured analysis process configuration library; wherein the preset structured analysis process configuration library is constructed in advance based on insight analysis processes corresponding to different task intents; extracting intent keywords from the insight analysis request based on the identified task intent, and performing semantic retrieval in a long-term memory vector library storing business terms and indicator definitions to obtain context information; and triggering the invocation of target tools matching the current step step by step according to the multi-step analysis instructions in the insight analysis execution path based on the context information, and using the execution result of the corresponding tool as the input for the corresponding subsequent step to obtain the insight analysis result.

[0083] Furthermore, this application also provides a computer-readable storage medium, which includes a stored program. When the program runs, it executes the insight analysis method based on the observation platform provided by the methods described above. This method includes: identifying the intent of a received insight analysis request; generating an insight analysis execution path containing multi-step analysis instructions based on the identified task intent and a pre-built structured analysis process configuration library; wherein the pre-built structured analysis process configuration library is constructed based on insight analysis processes corresponding to different task intents; extracting intent keywords from the insight analysis request based on the identified task intent, and performing semantic retrieval in a long-term memory vector library storing business terms and indicator definitions to obtain context information; and triggering the invocation of target tools matching the current step step by step according to the multi-step analysis instructions in the insight analysis execution path based on the context information, and using the execution result of the corresponding tool as the input for the corresponding subsequent step to obtain the insight analysis result.

[0084] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. 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.

[0085] 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.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application 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. Such 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 this application.

Claims

1. An insight analysis method based on an observation platform, characterized in that, include: The received insight analysis request is subjected to intent recognition, and based on the recognized task intent, combined with a preset structured analysis process configuration library, an insight analysis execution path containing multi-step analysis instructions is generated; wherein, the preset structured analysis process configuration library is constructed in advance based on insight analysis processes corresponding to different task intents; Based on the identified task intent, intent keywords are extracted from the insight analysis request, and semantic retrieval is performed in a long-term memory vector library that stores business terms and indicator definitions to obtain contextual information. Based on the context information, the target tool matching the current step is called step by step according to the multi-step analysis instructions in the insight analysis execution path, and the execution result of the corresponding tool is used as the input for the corresponding subsequent step to obtain the insight analysis result.

2. The insight analysis method based on an observation platform according to claim 1, characterized in that, Based on the context information, and according to the multi-step analysis instructions in the insight analysis execution path, the invocation of the target tool matching the current step is triggered step by step, including: For each step of the insight analysis execution path, the instruction intent of the corresponding step analysis instruction is identified, and based on the instruction intent and the context information, a preset tool library is searched to determine the target tool for the corresponding step analysis instruction. Based on the selected target tool, extract parameter entities from the insight analysis request and the context information, generate structured parameters that conform to the parameter format of the corresponding target tool, and configure the corresponding target tool.

3. The insight analysis method based on an observation platform according to claim 2, characterized in that, Based on the context information and the insight analysis execution path, a target tool is selected from the candidate tool set corresponding to the step analysis instruction, including: Based on the context information and the insight analysis execution path, and in conjunction with a preset selection strategy, a target tool is selected from the candidate tool set corresponding to the step analysis instruction; wherein, the preset selection strategy is used for: Determine the degree of fit between each candidate tool in the candidate toolset and the corresponding step analysis instructions to obtain the functional matching degree of each candidate tool; Based on each candidate tool in the candidate toolset, the insight analysis request, and the context information, determine whether the required parameters of the corresponding candidate tool can be completely extracted from the context information, and obtain the parameter compatibility of each candidate tool; Obtain the historical execution success rate and / or estimated execution cost of each candidate tool to obtain the reliability evaluation results of each candidate tool; Based on the functional matching degree of each candidate tool, the parameter compatibility of each candidate tool, and the reliability evaluation results of each candidate tool, a comprehensive evaluation is conducted on each candidate tool to determine the corresponding target tool.

4. The insight analysis method based on an observation platform according to claim 2, characterized in that, Based on the selected target tool, parameter entities are extracted from the insight analysis request and the context information to generate structured parameters that conform to the parameter format of the corresponding target tool, including: Based on the selected target tool, extract parameter entities from the insight analysis request and the context information; Based on the extracted parameter entities, matching nodes are searched in the pre-constructed business knowledge graph to obtain candidate nodes; wherein, the business knowledge graph is constructed in advance based on entities of different business data and the relationships between the entities, and each node of the business knowledge graph stores corresponding entity transformation rules configured in advance based on different tool parameter formats; Based on the candidate nodes, context disambiguation is performed in conjunction with the context information to obtain the target nodes, and the corresponding parameter entities are linked to the target nodes of the business knowledge graph. Based on the target node, the transformation rules for the corresponding target tool in the business knowledge graph are found, and the corresponding parameter entities are transformed according to the found transformation rules to obtain structured parameters that conform to the parameter format of the corresponding target tool.

5. The insight analysis method based on an observation platform according to claim 1, characterized in that, After extracting intent keywords from the insight analysis request based on the identified intent, and performing semantic retrieval in a long-memory vector library storing business terms and metric definitions to obtain contextual information, the process includes: The identified intent and the insight analysis request are matched and retrieved against the short-term memory that records the current conversation reasoning chain to obtain the historical interaction information corresponding to the dialogue round; The context information and the historical interaction information corresponding to the dialogue rounds are fused together.

6. The insight analysis method based on an observation platform according to claim 5, characterized in that, The context information and the historical interaction information corresponding to the dialogue rounds are fused, including: The context information is used as static knowledge, the historical interaction information corresponding to the dialogue rounds is used as dynamic dialogue context, and the attention weights of static knowledge and dynamic dialogue context are determined by combining the intent identified by the insight analysis request. Based on the static knowledge and the dynamic dialogue context, a weighted aggregation is performed using corresponding attention weights to obtain contextual information that integrates static knowledge and dynamic conversation context.

7. The insight analysis method based on an observation platform according to claim 1, characterized in that, Based on the identified task intent, and combined with a pre-defined structured analysis process configuration library, an insight analysis execution path containing multi-step analysis instructions is generated, including: Based on the identified intent, determine the task type and, in conjunction with the pre-set structured analysis process configuration library, obtain the corresponding insight analysis process configuration file; The insight analysis process configuration file is parsed to obtain a process topology structure containing multiple analysis step nodes, and a dynamic execution context for recording the execution status and intermediate results of each node is initialized according to the business parameters corresponding to the insight analysis request. Based on the dynamic execution context and preset optimization strategy, the process topology is dynamically adjusted to generate an insight analysis execution path; wherein, the dynamic adjustment includes pruning nodes that do not meet the execution conditions, attaching new analysis nodes, or selecting the optimal path from multiple executable branches.

8. The insight analysis method based on an observation platform according to claim 7, characterized in that, The process topology includes a directed acyclic graph, where each node encapsulates the corresponding step analysis instructions, input parameter specifications, and decision conditions for jumping to the next node. Based on the business parameters corresponding to the insight analysis request, initialize a dynamic execution context for recording the execution status and intermediate results of each node, including: The general business parameter entity is extracted from the insight analysis request, and the extracted general business parameter entity is parameter-bound to the starting node of the directed acyclic graph to initialize a dynamic execution context for recording the execution status and intermediate results of each node.

9. The insight analysis method based on an observation platform according to any one of claims 1 to 8, characterized in that, Intent identification is performed on received insight analysis requests, including: Load a preset agent configuration file and convert the contents of the preset agent configuration file into a structured object; wherein, the preset agent configuration file is configured in advance based on the role and responsibility of the agent; Key features are extracted from the structured object and filled into a preset system prompt template to generate a system prompt; Based on the system prompts, the received insight analysis request is subjected to intent recognition.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to execute the method of any one of claims 1 to 9 through the computer program.

Images