CN-121979902-A - Insight analysis method based on observation platform and electronic device

CN121979902ACN 121979902 ACN121979902 ACN 121979902ACN-121979902-A

Abstract

The application discloses an observation platform-based insight analysis method and an electronic device, and relates to the technical field of artificial intelligence, wherein the method comprises the steps of carrying out intention recognition on a received insight analysis request, and generating an insight analysis execution path containing multi-step analysis instructions according to recognized task intention and by combining a preset structured language analysis flow configuration library; according to the context information, analyzing the multi-step analysis instruction in the execution path according to the insights, triggering the call of the target tool matched with the current step in steps, and taking the execution result of the corresponding tool as the input of the corresponding subsequent step to obtain the insights analysis result. The method solves the defects of poor semantic understanding capability of the general intelligent agent and poor conversion accuracy of SQL sentences, and improves the semantic understanding capability and the data analysis efficiency of the intelligent agent.

Inventors

YANG MENG
YIN DESHUAI
ZHANG YUNHAO
HU BAICHUN

Assignees

青岛海尔科技有限公司
海尔优家智能科技（北京）有限公司
青岛海尔智能家电科技有限公司

Dates

Publication Date: 20260505
Application Date: 20251225

Claims (10)

1. An insight analysis method based on an observation platform, comprising: Performing intention recognition on the received insight analysis request, and generating an insight analysis execution path comprising a multi-step analysis instruction according to the recognized task intention and combining a preset structured language analysis flow configuration library, wherein the preset structured analysis flow configuration library is constructed based on different task intents in advance and corresponds to the insight analysis flow; Extracting intention keywords from the insight analysis request according to the identified task intention, and carrying out semantic retrieval in a long-term memory vector library stored with service terms and index definitions to obtain context information; According to the context information, triggering the calling of the target tool matched with the current step by step according to the multi-step analysis instruction in the insight analysis execution path, and taking the execution result of the corresponding tool as the input of the corresponding subsequent step to obtain the insight analysis result.
2. The observation platform-based insight analysis method of claim 1, wherein analyzing the multi-step analysis instructions in the execution path in the insight based upon the contextual information, step-wise triggering the invocation of the target tool matching the current step, includes: aiming at each step analysis instruction of the insight analysis execution path, respectively identifying instruction intentions corresponding to the step analysis instructions, searching a preset tool library based on the instruction intentions and the context information, and determining target tools corresponding to the step analysis instructions; and extracting parameter entities from the insight analysis request and the context information according to the selected target tool, generating structural parameters conforming to the parameter format of the corresponding target tool, and configuring the corresponding target tool.
3. The observation platform-based insight analysis method of claim 2, wherein selecting a target tool from a candidate tool set corresponding to the step analysis instructions based upon the contextual information and the insight analysis execution path, includes: Selecting a target tool from a candidate tool set corresponding to the step analysis instruction according to the context information and the insight analysis execution path and combining a preset selection strategy, wherein the preset selection strategy is used for: determining the fit degree of each candidate tool in the candidate tool set and the corresponding step analysis instruction to obtain the function matching degree of each candidate tool; determining whether parameters required by corresponding candidate tools can be completely extracted from the context information according to the candidate tools in the candidate tool set, the insight analysis request and the context information, and obtaining parameter compatibility of the candidate tools; Acquiring historical execution success rate and/or estimated execution cost of each candidate tool to obtain a reliability evaluation result of each candidate tool; and comprehensively evaluating each candidate tool according to the function matching degree of each candidate tool, the parameter compatibility of each candidate tool and the reliability evaluation result of each candidate tool, and determining a corresponding target tool.
4. The observation platform-based insight analysis method of claim 2, wherein extracting parameter entities from the insight analysis request and the context information in accordance with the selected target tool, generating structured parameters conforming to a corresponding target tool parameter format, includes: Extracting a parametric entity from the insight analysis request and the context information according to the selected target tool; Searching matched nodes in a pre-constructed business knowledge graph according to the extracted parameter entities to obtain candidate nodes, wherein the business knowledge graph is constructed based on the entities of different business data and the relation between the entities, and each node of the business knowledge graph stores entity conversion rules configured based on different tool parameter formats; According to the candidate nodes, carrying out context disambiguation by combining the context information to obtain target nodes, and linking corresponding parameter entities to the target nodes of the business knowledge graph; And searching a conversion rule of a corresponding target tool in the business knowledge graph according to the target node, and executing conversion on a corresponding parameter entity according to the searched conversion rule to obtain a structured parameter conforming to a parameter format of the corresponding target tool.
5. The observation platform-based insight analysis method of claim 1, wherein 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 index definitions, obtaining context information, comprising: Matching and searching the identified intention and the insight analysis request with a short-term memory bank for recording a current session reasoning chain to obtain dialogue round corresponding historical interaction information; And carrying out information fusion on the context information and the historical interaction information corresponding to the dialogue turns.
6. The observation platform-based insight analysis method of claim 5, wherein the information fusion of the context information and the dialog turn correspondence history interaction information includes: Taking the context information as static knowledge, taking the dialogue round corresponding history interaction information as dynamic dialogue context, and combining the insight analysis request corresponding identified intention to determine the attention weights of the static knowledge and the dynamic dialogue context; and carrying out weighted aggregation according to the static knowledge and the dynamic dialogue context and combining the corresponding attention weights to obtain the context information fusing the static knowledge and the dynamic dialogue context.
7. The observation platform-based insight analysis method of claim 1, wherein generating an insight analysis execution path including multi-step analysis instructions in combination with a pre-configured structured speech analysis flow configuration library based on the identified task intent, includes: determining task types according to the identified intents, and acquiring corresponding insight analysis flow configuration files by combining a preset structured speech analysis flow configuration library; Analyzing the insight analysis flow configuration file to obtain a flow topological structure comprising a plurality of analysis step nodes, initializing a dynamic execution context for recording the execution state and the intermediate result of each node according to the business parameters corresponding to the insight analysis request; And dynamically adjusting the flow topological structure according to the dynamic execution context and a preset optimization strategy to generate an insight analysis execution path, wherein the dynamic adjustment comprises cutting nodes which do not meet the execution conditions, mounting newly-added analysis nodes or selecting an optimal path from a plurality of executable branches.
8. The observation platform based insight analysis method of claim 7, wherein the flow topology includes a directed acyclic graph, each node in the directed acyclic graph encapsulating a corresponding step analysis instruction, an input parameter specification, and a decision condition for jumping to a next node; initializing a dynamic execution context for recording the execution state and the intermediate result of each node according to the corresponding service parameters of the insight analysis request, wherein the dynamic execution context comprises the following steps: And carrying out parameter binding on the extracted general service parameter entity and the initial node of the directed acyclic graph so as to initialize a dynamic execution context for recording the execution state and the intermediate result of each node.
9. The observation platform-based insight analysis method of any of claims 1 to 8, wherein intent recognition of the received insight analysis request includes: loading a preset agent configuration file and converting the content of the preset agent configuration file into a structural object, wherein the preset agent configuration file is configured based on roles and responsibilities of agents in advance; Extracting key features from the structured object, filling the extracted key features into a preset system prompt template, and generating a system prompt; And carrying out intention recognition on the received insight analysis request according to the system prompt.
10. An electronic device comprising a memory and a processor, characterized in that the memory has stored therein a computer program, the processor being arranged to execute the method according to any of claims 1 to 9 by means of the computer program.

Description

Insight analysis method based on observation platform and electronic device Technical Field The application relates to the technical field of artificial intelligence, in particular to an insight analysis method based on an observation platform and an electronic device. Background With the rapid development and maturation of large language model technology, artificial intelligence is evolving from single perception intelligence to cognition and decision intelligence, and artificial intelligence Agent (ARTIFICIAL INTELLIGENCE AGENT, AI-Agent for short) is becoming a key bridge connecting human intention and the digital world as a core product under this wave. The AI-Agent is an intelligent entity with autonomous understanding, planning, executing and feedback capabilities, and can receive natural language instructions of users, actively disassemble tasks, call tools (such as APIs, databases and code execution environments), and finally complete complex targets. At present, tracking the running condition of a service system is mainly realized through AI-Agent facing data analysis, by utilizing the powerful natural language understanding capability of a large language model, the service problem proposed by a user is resolved into structured intention and key entities, the resolved intention is converted into executable structured query language (Structured Query Language, SQL) sentences, the generated SQL sentences are submitted to a database for execution, and the returned result is presented to the user in a form of a table or a chart. However, in order to serve a wide service scenario, the model training data of the general-purpose AI-Agent is often universal, the service lines are more, a certain service or field cannot be focused, so that the Agent cannot accurately capture the real analysis intention of a user when facing a specific service term, an index caliber and complex service logic, the semantic understanding breadth and depth are insufficient, and the natural language is accurately mapped into SQL with correct grammar and correct logic, which is a challenging task, grammar or logic errors often occur in the existing generated SQL sentences, so that query failure or error results are caused, and the SQL conversion accuracy and robustness are poor. In addition, the capability of the AI-Agent analysis insight is weak, the active mining capability of the association relation behind the data is lacking, multi-dimensional drill-down, comparison and attribution analysis cannot be performed based on the query result, and the deep insight with commercial value is difficult to provide. Disclosure of Invention The application provides an insight analysis method and an electronic device based on an observation platform, which are used for solving the defects of poor semantic understanding capability of a general intelligent agent and poor conversion accuracy of SQL sentences in the prior art, improving the semantic understanding capability of the intelligent agent, effectively avoiding common logic divergence and reasoning loopholes of the general model and data ambiguity caused by unclear definition of manual understanding deviation or indexes, and accordingly disassembling a task which originally needs to generate extra-long complex SQL into a plurality of steps which only need to call specific tools, greatly reducing the probability of error generation of SQL sentences, realizing automatic execution of the whole process and improving the efficiency of data analysis. The application provides an insight analysis method based on an observation platform, which comprises the steps of carrying out intention recognition on a received insight analysis request, combining a preset structured language analysis flow configuration library according to recognized task intention to generate an insight analysis execution path containing multi-step analysis instructions, wherein the preset structured language analysis flow configuration library is constructed based on the insight analysis flow corresponding to different task intentions, extracting intention keywords from the insight analysis request according to the recognized task intention, carrying out semantic retrieval in a long-term memory vector library storing business terms and index definitions to obtain context information, triggering the invocation of a target tool matched with the current step according to the multi-step analysis instructions in the insight analysis execution path according to the context information, and taking the execution result of the corresponding tool as input corresponding to a later step to obtain the insight analysis result. According to the observation platform-based insight analysis method provided by the application, according to the context information, the multi-step analysis instruction in the execution path is analyzed according to the insight, and the call of the target tool matched with the current step is triggered step by