Retrieval enhancement generation system, method and equipment based on thinking library and medium

Abstract

The invention discloses a thinking library-based retrieval enhancement generation system, method and equipment and a medium. The thinking library-based retrieval enhancement generation system comprises a problem distillation module, a thinking library retrieval module, a knowledge retrieval module and a generation module, the problem distillation module is used for extracting context information from input problems and simplifying problem description; the thinking library retrieval module is used for storing and managing high-level thinking templates extracted from various tasks in a thinking library and retrieving related high-level thinking templates based on simplified problem description; the knowledge retrieval module is used for performing knowledge retrieval from a knowledge base according to the simplified problem description to generate documents or knowledge fragments; and the generation module is used for generating answers or contents based on the related high-level thinking templates, the documents or the knowledge fragments and the context information. According to the method, the reasoning enhancement technology and the dynamic information retrieval technology are combined, and the method can adapt to various complicated types of tasks and scenes.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and more specifically to a retrieval enhancement generation system, method, device and medium based on a thought library. Background Art

[0002] With the rapid development of artificial intelligence and natural language processing technologies, large-scale language models (LLMs) such as GPT-4 have performed well in text generation and understanding tasks. However, although these models have achieved remarkable results in various natural language tasks, when dealing with complex reasoning tasks in serious information service scenarios, they generate probabilistic error information that affects service quality and even business risks. In the face of these challenges, current technologies mainly rely on "retrieval-augmented generation" (RAG) and "high-level reasoning enhancement technology."

[0003] RAG is a method that combines the generative capabilities of language models and the retrieval capabilities of external knowledge bases to improve the relevance and accuracy of generated content. However, traditional RAG also has some inherent challenges:

[0004] Complexity of information integration: How to efficiently integrate the retrieved information with the generation process to ensure the consistency and relevance of the content is a key challenge.

[0005] Insufficient reasoning ability: Although RAG is good at acquiring and integrating information from the outside, it lacks the ability of deep reasoning and has difficulty handling tasks that require complex logical analysis.

[0006] Dependence on external knowledge: The effectiveness of RAG is highly dependent on the quality and coverage of the knowledge base. If the information in the knowledge base is insufficient or outdated, the quality of the generated results will be significantly affected.

[0007] In order to overcome the limitations of LLMs in complex reasoning tasks, researchers have proposed a series of high-level reasoning enhancement techniques, such as chain thinking and multiple query reasoning. These methods have played an important role in improving reasoning capabilities, but they also have obvious limitations:

[0008] High computational cost: These methods usually require multiple queries or complex reasoning path construction, resulting in high computational overhead and slow processing speed.

[0009] Poor task adaptability: Specific reasoning enhancement methods usually require carefully designed prompts or reasoning steps for the task, lacking generality and flexibility.

[0010] The limitations of existing technologies show that methods that rely solely on reasoning enhancement or retrieval enhancement are difficult to cope with complex and diverse tasks. In order to overcome the limitations of existing technologies when dealing with complex reasoning and generation tasks, it is necessary to combine high-level reasoning enhancement technology with dynamic information retrieval technology. Such a combination can not only obtain the latest information in the external knowledge base in real time and provide rich background support for the generation process, but also use high-level thinking templates to simplify the reasoning path of complex problems. Summary of the invention

[0011] In view of this, the present invention provides a retrieval enhancement generation system, method, device and medium based on a thought library, which combines reasoning enhancement technology with dynamic information retrieval technology and can adapt to various complex types of tasks and scenarios.

[0012] In order to achieve the above object, the present invention adopts the following technical solution:

[0013] A retrieval enhancement generation system based on a thinking library, comprising: a question distillation module, a thinking library retrieval module, a knowledge retrieval module and a generation module;

[0014] The question distillation module is used to extract context information from the input question and simplify the question description;

[0015] The thinking library retrieval module is used to store and manage high-level thinking templates extracted from various tasks in the thinking library and retrieve relevant high-level thinking templates based on simplified problem descriptions;

[0016] The knowledge retrieval module is used to retrieve knowledge from the knowledge base according to the simplified problem description and generate documents or knowledge fragments;

[0017] The generation module is used to generate answers or content based on relevant high-level thinking templates, documents or knowledge fragments, and context information.

[0018] Preferably, it also includes a dynamic adjustment and learning module for dynamically adjusting and optimizing the system according to the generated results and user feedback, including high-level thinking templates in the thinking library, retrieval strategies and knowledge in the knowledge base.

[0019] Preferably, the problem distillation module includes an input parsing unit, an information extraction unit and a problem simplification unit;

[0020] The input parsing unit is used to receive and parse questions input by users, and identify semantic information and keyword information;

[0021] The information extraction unit is used to extract context information based on semantic information and keyword information;

[0022] The problem simplification unit is used to decompose complex problems into simpler sub-problems based on context information.

[0023] Preferably, the thought library retrieval module includes a template storage unit, a template classification unit and a template retrieval unit;

[0024] The template storage unit is used to store high-level thinking templates extracted from various tasks in the thinking library;

[0025] The template classification unit is used to classify and index high-level thinking templates according to task type and complexity;

[0026] The template retrieval unit is used to retrieve relevant high-level thinking templates according to the simplified problem description.

[0027] Preferably, the knowledge retrieval module includes a query generation unit, a knowledge retrieval unit and a knowledge screening unit;

[0028] The query generation unit is used to generate a query for retrieval according to the simplified problem description;

[0029] The knowledge retrieval unit is used to retrieve documents or knowledge fragments matching the generated query from the knowledge base;

[0030] The knowledge screening unit is used to screen and filter documents or knowledge fragments.

[0031] Preferably, the generation module includes an information integration unit, a template application unit and an answer generation unit;

[0032] The information integration unit is used to integrate documents or knowledge fragments and context information;

[0033] The template application unit is used to form a reasoning path suitable for the current task based on the integrated information and the relevant high-level thinking template;

[0034] The answer generation unit is used to generate a final answer or content using LLMs according to the reasoning path of the current task.

[0035] Preferably, the dynamic adjustment and learning module includes a feedback collection unit, a template optimization unit, a strategy update unit and a knowledge update unit;

[0036] The feedback collection unit is used to collect user feedback on the generated result;

[0037] The template optimization unit is used to adjust and optimize the high-level thinking templates in the thinking library according to the feedback results;

[0038] The strategy updating unit is used to adjust and optimize the retrieval strategy according to the feedback result;

[0039] The knowledge updating unit is used to adjust and optimize the knowledge in the knowledge base according to the feedback results.

[0040] A retrieval enhancement generation method based on a thought library, comprising:

[0041] Extract contextual information from the input question to simplify the problem description;

[0042] Store and manage high-level thinking templates extracted from various tasks in the thinking library and retrieve relevant high-level thinking templates based on simplified problem descriptions;

[0043] Retrieve knowledge from the knowledge base based on the simplified problem description and generate documents or knowledge fragments;

[0044] Generate answers or content based on relevant high-level thinking templates, documents or knowledge fragments, and contextual information.

[0045] A computer device comprises: a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the steps of a retrieval enhancement generation method based on a thought library are implemented.

[0046] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a retrieval enhancement generation method based on a thought library.

[0047] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a retrieval enhancement generation system, method, device and medium based on a thought library, which has the following advantages:

[0048] Efficient reasoning ability: The LoT-RAG system of the present invention simplifies the reasoning process of complex problems by using high-level thinking templates. Compared with the traditional chain-of-thought (CoT) and multi-query reasoning methods (Multi-Query Reasoning), the present invention can construct an effective reasoning path in a shorter time, thereby improving the reasoning efficiency of the system.

[0049] Flexible task adaptability: It can adaptively select and apply appropriate thinking templates and retrieval strategies according to different task requirements. Compared with traditional chain-of-thought (CoT) and multi-query reasoning methods (Multi-Query Reasoning), it has stronger task adaptability. The system's thinking library and retrieval module can be dynamically adjusted to cope with various types of tasks and scenarios.

[0050] Enhanced generation quality: By combining dynamic information retrieval from external knowledge bases, the LoT-RAG system is able to generate high-quality answers with more context and details. The system is not only able to handle complex question-answering and content generation tasks, but also performs well in tasks involving deep background knowledge.

[0051] Continuous self-optimization capability: Through dynamic adjustment and learning modules, the LoT-RAG system can continuously optimize its thinking templates and retrieval strategies based on user feedback and generated results. The system will gradually improve its performance and adaptability as it handles more tasks.

[0052] Effective information integration: The LoT-RAG system performs well in information integration, efficiently combining retrieved information with reasoning templates to generate coherent and highly relevant content. The system's generation module can provide clear and logical answers in complex contexts. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0054] Figure 1 A schematic diagram of the structure of a retrieval enhancement generation system based on a thought library provided by the present invention.

[0055] Figure 2 A flow chart of a retrieval enhancement generation method based on a thought library provided by the present invention.

[0056] Figure 3 A flow chart of a retrieval enhancement generation method based on a thought library in another embodiment provided by the present invention. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] The embodiment of the present invention discloses a search enhancement generation system based on a thought library, such as Figure 1 As shown, it includes: question distillation module, thought base retrieval module, knowledge retrieval module and generation module;

[0059] The question distillation module is used to extract contextual information from the input question and simplify the question description;

[0060] The thinking library retrieval module is used to store and manage high-level thinking templates extracted from various tasks in the thinking library and retrieve relevant high-level thinking templates based on simplified problem descriptions;

[0061] The knowledge retrieval module is used to retrieve knowledge from the knowledge base according to the simplified problem description and generate documents or knowledge fragments;

[0062] The generation module is used to generate answers or content based on relevant high-level thinking templates, documents or knowledge fragments, and contextual information.

[0063] This system is a multi-round dialogue system, and the output of the previous round will serve as the historical dialogue information for the input of the next round.

[0064] In this embodiment, a dynamic adjustment and learning module is also included, which is used to dynamically adjust and optimize the system according to the generated results and user feedback, including high-level thinking templates in the thinking library, retrieval strategies and knowledge in the knowledge base.

[0065] In this embodiment, the question distillation module includes an input parsing unit, an information extraction unit, and a question simplification unit;

[0066] The input parsing unit is used to receive and parse questions input by users and identify semantic information and keyword information;

[0067] The information extraction unit is used to extract context information based on semantic information and keyword information, and represent it as structured data;

[0068] The problem simplification unit is used to decompose complex problems into simpler sub-problems based on contextual information.

[0069] In this embodiment, the thinking library retrieval module includes a template storage unit, a template classification unit and a template retrieval unit;

[0070] The template storage unit is used to store high-level thinking templates extracted from various tasks in the thinking library;

[0071] The template classification unit is used to classify and index high-level thinking templates according to task type and complexity;

[0072] The template retrieval unit is used to retrieve relevant high-level thinking templates based on the simplified problem description.

[0073] In this embodiment, the knowledge retrieval module includes a query generation unit, a knowledge retrieval unit and a knowledge screening unit;

[0074] The query generation unit is used to generate a query for retrieval based on the simplified problem description. For example, when answering a question about a historical event, the system will generate a query containing time and event keywords;

[0075] The knowledge retrieval unit is used to retrieve documents or knowledge fragments matching the generated query from the knowledge base, ensuring that the retrieved documents or knowledge fragments are highly relevant to the question;

[0076] The knowledge screening unit is used to screen and filter documents or knowledge fragments, retaining the most relevant documents or knowledge fragments for the generation process.

[0077] In this embodiment, the generation module includes an information integration unit, a template application unit, and an answer generation unit;

[0078] The information integration unit is used to integrate documents or knowledge fragments and context information to provide rich background support for the generation process;

[0079] The template application unit is used to form a reasoning path suitable for the current task based on the integrated information and the relevant high-level thinking template;

[0080] The answer generation unit is used to generate the final answer or content using LLMs according to the reasoning path of the current task, ensuring that the generated results are logically coherent and highly relevant.

[0081] In this embodiment, the dynamic adjustment and learning module includes a feedback collection unit, a template optimization unit, a strategy update unit and a knowledge update unit;

[0082] The feedback collection unit is used to obtain user feedback on the generated results;

[0083] The template optimization unit is used to adjust and optimize the high-level thinking templates in the thinking library according to the feedback results to improve the accuracy of reasoning and generation;

[0084] The strategy updating unit is used to adjust and optimize the retrieval strategy according to the feedback results to adapt to the changing task requirements;

[0085] The knowledge updating unit is used to adjust and optimize the knowledge in the knowledge base according to the feedback results, that is, to optimize the knowledge that has not been updated in time.

[0086] The following is an explanation based on specific cases:

[0087] Based on the diagnosis and treatment scenarios, a series of thinking templates for diagnosis and treatment related tasks are pre-defined, such as intelligent triage, auxiliary diagnosis, surgical assistance, etc.

[0088] When the user inputs a question, the question distillation module first parses the input, extracts relevant information, and simplifies the question. For example, if the user inputs the question "My feet are uncomfortable and it hurts when I walk. Which department should I go to?", the information is extracted {"Part of the disease": "Feet", "Description of symptoms": "It hurts when I walk"}, and the simplified question is: "Which department should I go to for foot pain?". Next, the thinking library retrieval module retrieves the relevant high-level thinking templates corresponding to the question based on the simplified question. For example, "Which department should I go to for foot pain?" belongs to the intelligent triage task, and the thinking template of the intelligent guidance task is "symptom analysis" -> "department knowledge base retrieval" -> "registration information query" -> "provide registration services" -> "generate answers", and then the system executes the steps in the thinking template step by step. The first step is "symptom analysis". The user's symptoms belong to foot surgery. The second step is "department knowledge base retrieval". The simplified question is generated by the query generation module to generate a query statement, and then the information of foot surgery is retrieved from the knowledge retrieval module, such as location information, X floor XX area, doctor information, etc. The third step is "registration information query". Further query the registration information of foot surgeons today from the knowledge retrieval module to see if the appointment is full. The fourth step is "providing registration service". Further knowledge screening is performed based on the query content in the third step. If the doctor has an appointment, the service link is obtained. If not, the service link is empty. The fifth step is "generating answers". According to the information obtained in the previous steps, the user's symptoms, department information, doctor information, and registration information are integrated with the user's questions, and then combined with the thinking template application, the final answer is generated by the generation module (LLMs), such as "According to your description, you should go to the foot surgery department, in the X floor X area. Currently, the department's doctor X has an appointment. Doctor X is a XXXX and is good at XXX (doctor information description). You can click on the service below to register directly and give comments and suggestions on this service." Finally, based on the generated results and user feedback, the evaluation suggestions will dynamically adjust and optimize the thinking templates, retrieval strategies and knowledge by dynamically adjusting the learning module.

[0089] This system can also be applied in the field of education, such as intelligent learning tutoring tasks. The following is an example of the processing steps of the system:

[0090] When the user inputs a question, the question distillation module first parses the input, extracts relevant information, and simplifies the question. The user inputs the question: "I am studying advanced mathematics and have encountered a problem with differential equations. I need help." Information extraction: {"Subject area": ​​"Advanced mathematics", "Problem type": "Differential equations"}, simplified question: "Solve the problem of differential equations in advanced mathematics". Next, based on the simplified question, the thinking template of the corresponding intelligent learning tutoring task is retrieved through the thinking library retrieval module. The thinking template of the intelligent learning tutoring task: "Knowledge point matching" -> "Learning resource retrieval" -> "Current user progress query" -> "Personalized tutoring plan generation" -> "Generate answers and tutoring material content". The system then executes the steps in the thinking template step by step. The first step is "knowledge point matching", which analyzes the type of differential equations in the user's questions, such as first-order differential equations, second-order differential equations, etc., and matches the corresponding knowledge points; the second step is "learning resource retrieval", which generates query statements through the query generation module after simplifying the problem, and then retrieves relevant teaching videos, handouts, exercises and problem-solving examples from the education resource library based on the knowledge point matching results; the third step is "current user progress query", which understands the user's learning progress; the fourth step is "personalized tutoring plan generation", which generates a personalized learning tutoring plan based on the user's learning progress and problem difficulty, including learning order, key content and difficulty analysis; information integration based on the information obtained in the previous steps combined with the user's questions , combined with the thinking template application generated by the generation module (LLMs) the final answer, such as "Based on your question, we have prepared a special learning package for advanced mathematics differential equations for you. First, you need to review the basic concepts and solving techniques of differential equations. Next, we recommend several related teaching videos and handouts for you, which cover the solution methods of first-order and second-order differential equations. In addition, we have matched you with a series of exercises and problem-solving examples to help you deepen your understanding. Please click on the learning path below to start your personalized learning journey, and complete the corresponding exercises to consolidate the knowledge points. If you encounter any problems during the learning process, you can always contact our online tutor for help." Finally, according to the user's learning progress and feedback, the thinking templates, retrieval strategies and educational resource libraries are dynamically adjusted and optimized through dynamic adjustment and learning modules to provide a learning experience that is more in line with user needs.

[0091] The embodiment of the present invention provides a retrieval enhancement generation method based on a thought library, such as Figure 2 As shown, including:

[0092] Extract contextual information from the input question to simplify the problem description;

[0093] Store and manage high-level thinking templates extracted from various tasks in the thinking library and retrieve relevant high-level thinking templates based on simplified problem descriptions;

[0094] Retrieve knowledge from the knowledge base based on the simplified problem description and generate documents or knowledge fragments;

[0095] Generate answers or content based on relevant high-level thinking templates, documents or knowledge fragments, and contextual information.

[0096] In this embodiment, if Figure 3 As shown, it also includes dynamically adjusting and optimizing the system according to the generated results and user feedback, including high-level thinking templates in the thinking library, retrieval strategies, and knowledge in the knowledge base.

[0097] In this embodiment, context information is extracted from the input question to simplify the question description, specifically including:

[0098] Receive and parse user input questions, and identify semantic information and keyword information;

[0099] Extract context information based on semantic information and keyword information and represent it as structured data;

[0100] Decompose complex problems into simpler sub-problems based on contextual information.

[0101] In this embodiment, high-level thinking templates extracted from various tasks are stored and managed in the thinking library, and relevant high-level thinking templates are retrieved based on the simplified problem description, specifically including:

[0102] Store high-level thinking templates extracted from various tasks in the thinking library;

[0103] Categorize and index high-level thinking templates according to task type and complexity;

[0104] Retrieve relevant high-level thinking templates based on simplified problem description.

[0105] In this embodiment, knowledge retrieval is performed from the knowledge base according to the simplified problem description to generate documents or knowledge fragments, specifically including:

[0106] Generate a query for retrieval based on a simplified problem description. For example, when answering a question about a historical event, the system will generate a query containing time and event keywords.

[0107] Retrieve documents or knowledge fragments matching the generated query from the knowledge base, ensuring that the retrieved documents or knowledge fragments are highly relevant to the question;

[0108] Screen and filter the documents or knowledge fragments, and retain the most relevant documents or knowledge fragments for the generation process.

[0109] In this embodiment, the answer or content is generated based on the relevant high-level thinking template, document or knowledge fragment, and context information, specifically including:

[0110] Integrate documents or knowledge fragments and contextual information to provide rich background support for the generation process;

[0111] Based on the integrated information and according to the relevant high-level thinking templates, form a reasoning path suitable for the current task;

[0112] Based on the reasoning path of the current task, LLMs are used to generate the final answer or content, ensuring that the generated results are logically coherent and highly relevant.

[0113] In this embodiment, the system is dynamically adjusted and optimized according to the generated results and user feedback, including high-level thinking templates in the thinking library, retrieval strategies and knowledge in the knowledge base, specifically including:

[0114] Get user feedback on generated results;

[0115] Adjust and optimize high-level thinking templates in the thinking library based on feedback results to improve the accuracy of reasoning and generation;

[0116] Adjust and optimize retrieval strategies based on feedback results to adapt to changing task requirements;

[0117] Adjust and optimize the knowledge in the knowledge base based on the feedback results, that is, optimize the knowledge that has not been updated in time.

[0118] This embodiment provides a computer device, including: a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the steps of a retrieval enhancement generation method based on a thought library are implemented.

[0119] This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of a retrieval enhancement generation method based on a thought library are implemented.

[0120] A person skilled in the art can understand that: all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, etc. Various media that can store program codes.

[0121] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0122] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A retrieval enhancement generation system based on thought library, It is characterized in that include: Question distillation module, thought base retrieval module, knowledge retrieval module and generation module; The question distillation module is used to extract context information from the input question and simplify the question description; The thinking library retrieval module is used to store and manage high-level thinking templates extracted from various tasks in the thinking library and retrieve relevant high-level thinking templates based on simplified problem descriptions; The knowledge retrieval module is used to retrieve knowledge from the knowledge base according to the simplified problem description and generate documents or knowledge fragments; The generation module is used to generate answers or content based on relevant high-level thinking templates, documents or knowledge fragments, and context information.

2. A search enhancement generation system based on a thought library according to claim 1, It is characterized in that It also includes a dynamic adjustment and learning module, which is used to dynamically adjust and optimize the system according to the generated results and user feedback, including high-level thinking templates in the thinking library, retrieval strategies and knowledge in the knowledge base.

3. A search enhancement generation system based on a thought library according to claim 1, It is characterized in that The question distillation module includes an input parsing unit, an information extraction unit and a question simplification unit; The input parsing unit is used to receive and parse questions input by users, and identify semantic information and keyword information; The information extraction unit is used to extract context information based on semantic information and keyword information; The problem simplification unit is used to decompose complex problems into simpler sub-problems based on context information.

4. A search enhancement generation system based on a thought library according to claim 1, It is characterized in that The thought library retrieval module includes a template storage unit, a template classification unit and a template retrieval unit; The template storage unit is used to store high-level thinking templates extracted from various tasks in the thinking library; The template classification unit is used to classify and index high-level thinking templates according to task type and complexity; The template retrieval unit is used to retrieve relevant high-level thinking templates according to the simplified problem description.

5. A search enhancement generation system based on a thought library according to claim 1, It is characterized in that The knowledge retrieval module includes a query generation unit, a knowledge retrieval unit and a knowledge screening unit; The query generation unit is used to generate a query for retrieval according to the simplified problem description; The knowledge retrieval unit is used to retrieve documents or knowledge fragments matching the generated query from the knowledge base; The knowledge screening unit is used to screen and filter documents or knowledge fragments.

6. A search enhancement generation system based on a thought library according to claim 1, It is characterized in that The generation module includes an information integration unit, a template application unit and an answer generation unit; The information integration unit is used to integrate documents or knowledge fragments and context information; The template application unit is used to form a reasoning path suitable for the current task based on the integrated information and the relevant high-level thinking template; The answer generation unit is used to generate a final answer or content using LLMs according to the reasoning path of the current task.

7. A search enhancement generation system based on a thought library according to claim 2, It is characterized in that The dynamic adjustment and learning module includes a feedback collection unit, a template optimization unit, a strategy update unit and a knowledge update unit; The feedback collection unit is used to obtain user feedback on the generated result; The template optimization unit is used to adjust and optimize the high-level thinking templates in the thinking library according to the feedback results; The strategy updating unit is used to adjust and optimize the retrieval strategy according to the feedback result; The knowledge updating unit is used to adjust and optimize the knowledge in the knowledge base according to the feedback results.

8. A retrieval enhancement generation method based on thought library, It is characterized in that include: Extract contextual information from the input question to simplify the problem description; Store and manage high-level thinking templates extracted from various tasks in the thinking library and retrieve relevant high-level thinking templates based on simplified problem descriptions; Retrieve knowledge from the knowledge base based on the simplified problem description and generate documents or knowledge fragments; Generate answers or content based on relevant high-level thinking templates, documents or knowledge fragments, and contextual information.

9. A computer device, It is characterized in that include: A memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the steps of a retrieval enhancement generation method based on a thought library are implemented.

10. A computer-readable storage medium, It is characterized in that The storage medium stores a computer program, which, when executed by a processor, implements the steps of a retrieval enhancement generation method based on a thought library.

Images