Deep search method and system based on large language model

Abstract

The invention discloses a deep search method and system based on a large language model. On the basis of initial query of the user, a targeted question is generated through a deep question-asking agent to perform deep question-asking on the user; obtaining an answer of the user to the deep question to form a complete query context; based on user query and answer, accurate search keywords are generated through a search keyword agent; executing a search operation, and obtaining and preliminarily analyzing a search result; and performing comprehensive analysis on all search results through the summarizing agent to generate a structured and readable final answer. According to the invention, through cooperative work of a plurality of specialized agents, search behaviors of human experts are simulated, and deep understanding of user information requirements, multi-round iterative search and intelligent result integration are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, in particular to a deep search method and system based on large language models, which is used for intelligent information retrieval, analysis and integration in massive Internet information to meet users' complex information needs. BACKGROUND

[0002] With the rapid development of the Internet, the amount of information is growing explosively, and the results obtained by users entering keywords in search engines often contain a large amount of redundant, repetitive and even incorrect information. Traditional search engines are mainly based on keyword matching and sorting algorithms, lacking deep understanding of users' real needs and intelligent integration capabilities of search results.

[0003] 1. Users have difficulty accurately expressing complex information needs, resulting in low search efficiency.

[0004] 2. The search results lack semantic understanding and intelligent filtering, and users need to manually filter a large amount of information.

[0005] 3. It is difficult to implement multi-round iterative search and handle complex queries that require multi-dimensional information integration.

[0006] 4. The search results are presented in a single form, lacking structured organization and deep analysis. SUMMARY

[0007] To solve the technical problems existing in the prior art, the present application provides a deep search method and system based on large language models, which simulates the search behavior of human experts through the cooperative work of multiple specialized intelligent agents, realizes the deep understanding of users' information needs, multi-round iterative search and intelligent result integration.

[0008] The technical solution of the present application is as follows: In a first aspect, the present application provides a deep search method based on large language models, comprising the following steps: Deep questioning stage: receiving the user's initial query, generating targeted questions through a deep questioning intelligent agent to conduct deep questioning on the user, and guiding the user to clarify the real information needs.

[0009] Demand clarification stage: obtaining the user's answers to the deep questioning, and forming a complete query context.

[0010] Search keyword generation stage: generating accurate search keywords based on the user's query and answers through a search keyword intelligent agent.

[0011] Iterative deep search stage: performing search operations, obtaining and preliminarily analyzing search results.

[0012] Result integration and summary phase: the agent analyzes all search results and generates a structured and readable final answer.

[0013] In one possible implementation, the depth questioning agent receives the user's initial query, analyzes the ambiguous or incomplete parts of the query based on preset instructions and current date information, and generates targeted questions accordingly.

[0014] In one possible implementation, the iterative deep search phase specifically includes: (a) Perform searches in parallel for each generated keyword.

[0015] (b) Obtain and preliminarily analyze search results through the content search agent.

[0016] (c) Based on the current search results, assess whether the information is sufficient.

[0017] (d) If the information is not sufficient, generate new search keywords for the next round of search.

[0018] In one possible implementation, the three-dimensional instantiation mechanism of "instruction semantic guidance - function module empowerment - context state constraint" is formed by configuring the instruction set containing task semantic guidance and output constraints for the basic large language model, grafting the plug-in module with domain-specific functions, and binding the environment variables containing the user query context, historical interaction state, and domain knowledge, thereby obtaining an intelligent agent with specific functions.

[0019] In one possible implementation, the method further includes a progress tracking and error handling mechanism, specifically as follows: Progress tracking: the system records and displays the current progress before and after each key step is executed.

[0020] Error handling: capture and handle possible errors.

[0021] Log recording: record detailed execution logs for problem diagnosis and system optimization.

[0022] In one possible implementation, possible errors include API call failure, agent response exception, user input exception, and empty search results.

[0023] In a second aspect, the embodiments of the present application provide a deep search system based on a large language model, which includes multiple specialized intelligent agent modules, each module being constructed based on a large language model and having specific instructions and functions: Depth questioning agent: responsible for receiving the user's initial query, generating depth questions, and helping to clarify user requirements.

[0024] Search keyword intelligent agent: responsible for generating accurate search keywords based on user queries and answers.

[0025] Content search intelligent agent: responsible for performing searches in parallel for each keyword generated, obtaining and preliminarily analyzing search results.

[0026] Summary intelligent agent: responsible for comprehensive analysis and integration of all search results, generating structured and easy-to-read final answers.

[0027] In one possible implementation, the system further includes a workflow manager responsible for coordinating workflows between intelligent agents, managing search iteration processes, and providing progress tracking and error handling mechanisms.

[0028] In a third aspect, the embodiments of the present application provide an electronic device, comprising a processor and a memory.

[0029] The memory is used to store a computer program.

[0030] The processor is used to execute the program stored on the memory, and implement the deep search method described in any of the embodiments of the present application.

[0031] In a fourth aspect, the embodiments of the present application provide a computer readable storage medium, which stores a computer program. When the computer program is executed by a processor, the deep search method described in any of the embodiments of the present application is implemented.

[0032] In a fifth aspect, the embodiments of the present application provide a computer program product containing instructions, which, when executed on a computer, cause the computer to perform the deep search method described in any of the embodiments of the present application.

[0033] The beneficial effects of the present application are as follows: 1. Improve search accuracy: through the deep questioning mechanism, accurately understand the real needs of users, reduce invalid search; deep questioning mechanism: based on the initial query of the user, the deep questioning intelligent agent combines the preset instructions and the current date information, accurately analyzes the fuzzy and incomplete parts in the user query, generates targeted questions, guides the user to clarify the real needs, solves the problem that traditional search is difficult to understand the user's fuzzy intention. 2. Enhance search depth: through multiple rounds of iterative search, realize in-depth exploration of complex problems; multiple rounds of iterative deep search: set the maximum number of search iterations, search keywords in parallel through the content search intelligent agent, and evaluate the sufficiency of the search results in real time according to the search results, dynamically generate new keywords for the next round of search, break through the limitations of traditional single search, and deeply mine complex information. 3. Improve information quality: filter low-quality information through intelligent screening and analysis, and provide high-value content; precise keyword generation strategy: after the search keyword intelligent agent receives the complete query context, it first determines whether the existing data is sufficient to answer the question, and if not, it generates independent and complete keywords without pronouns and separated by semicolons to improve search accuracy. 4. Optimize user experience: provide structured and easy-to-read search results to reduce user information processing burden; intelligent result integration: summarize the user query, answer and all search results of the intelligent agent, conduct deep analysis based on preset instructions, and generate structured and easy-to-read final answers to change the traditional search result fragmentation status.

[0034] 5. Adapt to complex queries: capable of handling complex query requirements that require multi-dimensional information integration.

[0035] 6. Improve search efficiency: reduce user manual search time through parallel search and intelligent processing; collaborative work and management architecture: the system is composed of four intelligent agents of deep questioning, search keywords, content search and summary, and a workflow manager; the workflow manager coordinates the workflow of each intelligent agent, manages search iterations, provides progress tracking and error handling, and ensures efficient and stable operation of the system. BRIEF DESCRIPTION OF DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other embodiments can also be obtained by those skilled in the art based on these drawings.

[0037] Figure 1 The overall flowchart of the deep search method based on the large language model of the present application.

[0038] Figure 2 The flowchart of the deep questioning stage of the present application.

[0039] Figure 3 The flowchart of the iterative deep search stage of the present application.

[0040] Figure 4 The architecture diagram of the multi-agent collaborative work in the present application. Figure 5 The flowchart of the embodiment of the present application for converting the base model into an intelligent agent instance with specific functions. DETAILED DESCRIPTION

[0041] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art based on the present application belong to the scope of protection of the present application.

[0042] The embodiment of the present application provides a deep search system based on a large language model, which mainly consists of four core agents, each of which is constructed based on a large language model and is configured with specific instructions and functions. Deep question agent: based on the user's initial query, generate targeted deep questions to help clarify the user's real needs.

[0043] Search keyword agent: based on user queries and answers, generate accurate search keywords.

[0044] Content search agent: perform search operations on each generated keyword in parallel, and obtain and preliminarily analyze the search results.

[0045] Summary agent: comprehensively analyze and integrate all search results to generate structured and easy-to-read final answers.

[0046] The system also includes a workflow manager responsible for coordinating the workflow between the agents, managing the search iteration process, and providing progress tracking and error handling mechanisms.

[0047] In one possible implementation, by configuring the basic large language model with a set of instructions containing task semantic guidance and output constraints, grafting plug-in modules with domain-specific functions (such as requirement analyzers, term extraction engines, etc.), and binding environment variables containing user query context, historical interaction state, and domain knowledge, a three-dimensional instantiation mechanism of "instruction semantic guidance - function module empowerment - context state constraint" is formed. Among them, instruction configuration defines the task target and output specification of the agent, function plug-in gives special abilities such as data processing and tool calling, and context environment dynamically injects scene information and state parameters. The three work together to transform the basic model into an agent instance with specific functions, realizing the role differentiation from general model to specialized task execution unit.

[0048] In one possible implementation, the instruction configuration is shown in Table 1, and the function plug-in is shown in Table 2: Table 1. Instruction configuration Table 2. Function plug-in The embodiment of the present application also provides a deep search method based on a large language model, including the following steps: Step (1) Depth Questioning Phase: Depth questioning is one of the key innovations of the invention, its purpose is to help users clarify the real information needs. The specific implementation steps are as follows: Step (1-1) The system receives the user's initial query; Step (1-2) Pass the query to the depth questioning agent; Step (1-3) The depth questioning agent analyzes the ambiguous or incomplete parts in the user query based on the preset instructions and current date information; Step (1-4) Based on the ambiguous or incomplete parts in the user query, generate targeted questions that are direct and valuable and can help clarify the user's real needs; Step (1-5) Return the generated depth questions to the user and wait for the user to answer.

[0049] Step (2) Demand Clarification Phase: In this phase, the system obtains the user's answers to the depth questions, forming a complete query context: Step (2-1) The system receives the user's answers to the depth questions; Step (2-2) Combine the user's initial query and answers to form a complete query context; Step (2-3) Pass the query context to the next stage of the search keyword agent.

[0050] Step (3) Search Keyword Generation Phase Search keyword generation is a key step to achieve accurate search, the specific implementation is as follows: Step (3-1) The search keyword agent receives the complete query context; Step (3-2) Based on the preset instructions, analyze the user's needs and the current known information; Step (3-3) Determine whether the current information is sufficient to answer the user's question: a. If sufficient, output "no need to search"; b. If not enough, generate search keywords; Step (3-4) The keywords should be independent and complete, avoid using pronouns, and use semicolons to separate.

[0051] Step (4) Iterative Deep Search Phase: Iterative deep search is the core function of the invention, through multiple rounds of search to continuously explore information related to the user's needs: Step (4-1) The system sets the maximum number of search iterations (such as 5 times); Step (4-2) For each round of search: (a) Get the search keywords for the current round; (b) Perform search for each keyword in parallel: Call content search agent to perform search; get and record search results.

[0052] (c) Aggregate search results for all keywords in current iteration; (d) Update known information set; (e) Call search keyword agent again to determine whether to continue search based on updated information: If output is "no need to search", end search iteration; if new keywords are generated, proceed to next iteration of search; Step (4-3) End search iteration when maximum number of iterations is reached or search keyword agent determines that information is sufficient.

[0053] Content search agent is responsible for performing actual search operation, which can call external search API and other private search tools (such as Exa search API) to obtain information.

[0054] Step (5) Result integration and summary stage: After completing all search iterations, the system needs to integrate and summarize the collected information: Step (5-1) Summary agent receives initial user query, user answers and all search results.

[0055] Step (5-2) Based on preset instructions, comprehensive analysis is performed on all information.

[0056] Step (5-3) Generate structured and easy-to-read final answer to ensure that the content is accurate, comprehensive and meets user needs; Step (5-4) Return the final answer to the user.

[0057] In one possible implementation, to improve system reliability and user experience, the embodiment of the present application further includes a progress tracking and error handling mechanism, which is as follows: Step (6-1) Progress tracking: before and after the execution of each key step, the system records and displays the current progress, including: (a) Depth question generation.

[0058] (b) Search iteration round.

[0059] (c) Keyword generation.

[0060] (d) Search execution status.

[0061] (e) Result integration status.

[0062] Step (6-2) Error Handling: The system captures and handles possible errors, including API call failures, agent response abnormalities, user input abnormalities, and empty search results.

[0063] (1) API call failure, the processing flow is as follows: The large model generates parameter adjustment schemes according to error types (such as replacing API keys, adjusting request headers).

[0064] When multiple failures occur (exceeding the set threshold), call the large model to generate backup search strategies (such as using synonyms to replace keywords).

[0065] (2) Agent response abnormality, the processing flow is as follows: Format exception: The large model generates format correction prompts according to the pre-set Schema.

[0066] Semantic exception: Use the large model to regenerate the response.

[0067] (3) User input abnormality, the processing flow is as follows: Mild abnormality (such as irregular format): The large model generates automatic correction prompts.

[0068] Severe abnormality (such as semantic irrelevance): Trigger the deep questioning agent to regenerate the follow-up questions.

[0069] Special scenario: When the user input contains sensitive content, the large model generates compliance prompts and terminates the process.

[0070] (4) Search results are empty, the processing flow is as follows: Analyze the reason for empty results: Determine whether the keyword problem or resource deficiency through the large model.

[0071] Keyword optimization: Generate synonyms, expanded words, or context-related words.

[0072] When multiple empty results occur (exceeding the set threshold), trigger the backtracking mechanism to return to the deep questioning stage to re-clarify the requirements.

[0073] Step (6-3) Log Recording: The system records detailed execution logs for problem diagnosis and system optimization.

[0074] The present application is applicable to various complex information retrieval scenarios, including but not limited to: 1. Academic research: Helps researchers quickly obtain the latest research results and related literature in a specific field.

[0075] 2. Market research: Assists enterprises in analyzing market trends, competitor conditions, and user needs.

[0076] 3. Technical Consultation: Providing in-depth technical problem analysis and solutions for technical personnel.

[0077] 4. Medical Consultation: Assisting medical professionals in obtaining the latest medical research and treatment plans.

[0078] 5. Legal Consultation: Helping legal practitioners retrieve relevant regulations and cases.

[0079] 6. Educational Assistance: Providing in-depth learning materials and knowledge integration for students and teachers.

[0080] In one possible implementation, the embodiments of the present invention can be optimized and extended according to specific application requirements: 1. Agent Model Selection: Different sizes and capabilities of large language models can be selected according to task complexity and response speed requirements.

[0081] 2. Search API Integration: Multiple search APIs such as Google, Bing, and professional databases can be integrated.

[0082] 3. Domain Adaptation: By adjusting agent instructions and prompt words, the system can adapt to specific domain professional search requirements.

[0083] 4. Multi-modal Extension: The system can support search and analysis of multi-modal information such as images and videos.

[0084] 5. User Preference Learning: The system can incorporate user feedback mechanisms to learn and adapt to user search preferences.

[0085] Through the above detailed description, the present invention provides a deep search method based on large language models, which can effectively solve the limitations of traditional search methods in handling complex information needs, improve search accuracy and information quality, and provide users with a more intelligent information retrieval experience. The following innovations are included: 1. Multi-agent collaborative deep search method: covering deep questioning, demand clarification, keyword generation, iterative deep search, result integration and summary, etc. Complete steps, through professional agent simulation of human expert search behavior, realize the deep understanding of user information needs and accurate search. 2. Deep questioning agent working mode: based on user initial query, preset instructions and date information to generate targeted questions to assist in clarifying user demand. 3. Search keyword agent judgment and generation logic: judging the sufficiency of the query context to generate accurate search keywords. 4. Iterative deep search execution process: including setting the maximum number of iterations, keyword parallel search, dynamic evaluation based on results and new keyword generation in multiple rounds of search process. 5. Result integration and summary agent processing method: comprehensive analysis of multiple information, structured and easy-to-read final answer processing strategy. 6. Deep search system based on large language model: composed of multiple specialized agent modules with specific instructions and functions and workflow managers, a system architecture that realizes collaborative work and efficient management. 7. Progress tracking and error handling mechanism: progress record display for key steps of the system, capture and processing of various abnormal conditions, and detailed log recording guarantee mechanism.

[0086] The embodiments of the present application also provide an electronic device. The embodiments of the present application provide an electronic device, which includes a processor and a memory.

[0087] The memory is configured to store a computer program.

[0088] The processor is configured to execute the program stored in the memory to implement any method described in the present application.

[0089] In a possible implementation, the electronic device of the embodiments of the present application further includes a communication interface and a communication bus, wherein the processor, the communication interface, and the memory complete mutual communication through the communication bus.

[0090] The communication bus mentioned in the above electronic device can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or only one type of bus.

[0091] The communication interface is configured to communicate between the above electronic device and other devices.

[0092] The memory can include a random access memory (RAM) and can also include a non-volatile memory (NVM), for example, at least one disk memory. Optionally, the memory can also be at least one storage device located away from the aforementioned processor.

[0093] The processor described above can be a general processor, including a central processing unit (CPU), a network processor (NP), etc.; can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.

[0094] In another embodiment provided in the present application, a computer readable storage medium is provided, and the computer readable storage medium stores a computer program. The computer program is executed by a processor to implement the method described in any of the embodiments of the present application.

[0095] In another embodiment provided in the present application, a computer program product is provided, and the computer program product includes instructions. When the computer program product is executed on a computer, the computer is caused to perform the method described in any of the embodiments of the present application.

[0096] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable device. The computer instructions can be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another computer readable storage medium, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) mode. The computer readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. integrated with one or more available media. The available medium can be a magnetic medium (for example, floppy disk, hard disk, magnetic tape), an optical medium (for example, DVD), or a semiconductor medium (for example, solid state disk (SSD)) and the like.

[0097] It is to be noted that, as used in this document, the term "thick" refers to a thickness of a layer of material, and the term "thin" refers to a thickness of a layer of material that is less than the thickness of the layer of material referred to as "thick". It is to be understood that the terms "thick" and "thin" are not intended to be limiting, and that the thickness of a layer of material can be any thickness, and that the thickness of a layer of material can be the same as, or different from, the thickness of another layer of material. It is to be understood that, as used in this document, the terms "first", "second", "third", etc., are used only to distinguish one element or operation from another, and do not necessarily imply any actual relationship or order between such elements or operations. Moreover, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the presence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0098] Each of the embodiments in the present specification is described in a related manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts between the embodiments can be understood by referring to each other.

[0099] The above only describes the preferred embodiments of the present application, and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A deep search method based on a large language model, characterized in that: The following steps are involved: In-depth questioning stage: Receive the user's initial query, generate targeted questions through the in-depth questioning agent, and ask the user in-depth questions to guide the user to clarify their real information needs; Demand clarification stage: Obtain the user's answers to in-depth questions to form a complete query context; Search keyword generation phase: Based on user queries and answers, the search keyword agent generates accurate search keywords; Iterative deep search phase: perform search operations, obtain and preliminarily analyze search results; Result integration and summary stage: All search results are comprehensively analyzed by the summary agent to generate a structured and easy-to-read final answer.

2. A deep search method based on a large language model according to claim 1, characterized in that: The deep questioning agent receives the user's initial query and analyzes the vague or incomplete parts of the user's query based on preset instructions and current date information to generate targeted questions.

3. The deep search method based on a large language model according to claim 1, characterized in that: The iterative depth search phase specifically includes: (a) Perform searches in parallel for each generated keyword; (b) Obtaining and preliminarily analyzing search results through content search agents; (c) assessing the sufficiency of information based on the current search results; (d) If the information is insufficient, new search keywords are generated and the next round of search is carried out.

4. A deep search method based on a large language model according to claim 1, characterized in that: By configuring the basic large language model with an instruction set that includes task semantic guidance and output constraints, grafting plug-in modules with domain-specific functions, and binding environment variables containing user query context, historical interaction status, and domain knowledge, a three-dimensional instantiation mechanism of "instruction semantic guidance - functional module empowerment - contextual state constraints" is formed, thereby obtaining an intelligent agent with specific functions.

5. The deep search method based on a large language model according to claim 1, characterized in that: The method also includes progress tracking and error handling mechanisms, as follows: Progress tracking: Before and after each key step, the system records and displays the current progress; Error handling: Capture and handle possible errors; Logging: Record detailed execution logs to facilitate problem diagnosis and system optimization.

6. A deep search method based on a large language model according to claim 5, characterized in that: Possible errors include: API call failure, abnormal agent response, abnormal user input, and empty search results.

7. A deep search system based on a large language model, characterized in that: It includes multiple specialized agent modules, each built on a large language model with specific instructions and functions: Deep Questioning Agent: Responsible for receiving the user's initial query, generating deep questions, and helping to clarify user needs; Search keyword agent: responsible for generating accurate search keywords based on user queries and answers; Content search agent: responsible for executing searches for each generated keyword in parallel, obtaining and preliminarily analyzing search results; Summarizing agent: responsible for comprehensive analysis and integration of all search results to generate a structured and easy-to-read final answer.

8. A deep search system based on a large language model according to claim 7, characterized in that: The system also includes a workflow manager, which is responsible for coordinating the workflow between agents, managing the search iteration process, and providing progress tracking and error handling mechanisms.

9. An electronic device, characterized in that: including processor and memory; The memory is used to store computer programs; The processor is configured to implement the depth search method according to any one of claims 1 to 6 when executing a program stored in the memory.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the depth search method according to any one of claims 1 to 6 is implemented.

Images