Symbolic logic reasoning optimization system based on large language model

Abstract

This invention discloses a symbolic logic reasoning optimization system for a large language model, comprising three core modules: a hierarchical prompt generator, a chained reasoning engine, and a symbolic expression optimizer. It achieves an end-to-end processing flow from scene input to optimization result output. The system first constructs an initial symbolic feature space based on user-input scene information. The hierarchical prompt generator generates scene initialization prompts, thought chain guidance prompts, and feedback integration prompts based on the symbolic feature space and preset scene prompt templates. The chained reasoning engine constructs and optimizes a population of symbolic expressions based on the symbolic feature space. It then inputs the current context information into the large language model using thought chain guidance prompts and feedback integration prompts, prompting the model to provide optimization suggestions. Finally, the symbolic expression optimizer transforms the reasoning results output by the large language model into executable optimization operations, iterating continuously until a termination condition is met.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence and symbolic logic reasoning technology, and specifically refers to a symbolic logic reasoning optimization system based on a large language model. Background Technology

[0002] In recent years, symbolic logic reasoning technology has been widely used in the field of artificial intelligence and has become an important way to achieve explainable artificial intelligence. Existing technologies mainly develop along two routes: one is rule systems based on expert knowledge, which realize logical deduction by manually defining reasoning rules. Although this method has strong interpretability, the rule design is cumbersome and it is difficult to cover complex scenarios; the other is reasoning methods based on machine learning, which learn reasoning patterns through data-driven approaches. Although this method has strong adaptability, it often lacks logical rigor and interpretability.

[0003] Current technological development faces several key challenges: traditional symbolic reasoning systems perform poorly when dealing with fuzzy logic expressed in natural language; the search space in complex scenarios is huge, leading to low reasoning efficiency; existing methods struggle to effectively balance the rigor and flexibility of reasoning; and there is a lack of effective mechanisms to integrate domain knowledge and common-sense reasoning capabilities. Summary of the Invention

[0004] The main objective of this invention is to provide a symbolic logic reasoning optimization system based on a large language model, which solves the problems existing in the prior art. It focuses on enhancing the optimization capability of the symbolic logic reasoning system by utilizing a large language model. By integrating the semantic understanding capability of the large language model with the rigor of the traditional symbolic reasoning system, it achieves a more efficient and intelligent logical reasoning process.

[0005] To achieve the above objectives, the solution of the present invention is:

[0006] A symbolic logic reasoning optimization system based on a large language model comprises three core modules: a hierarchical prompt generator, a chained reasoning engine, and a symbolic expression optimizer. It implements an end-to-end processing flow from scene input to optimization result output. The system first constructs an initial symbolic feature space based on the scene information input by the user. The three core modules function as follows:

[0007] The multi-level prompting architecture of the layered prompt generator includes scene initialization prompts. Mind chain guidance prompts Integration of feedback prompts First, the hierarchical hint generator will use the symbol feature space. The information is combined with the preset scene prompt template to generate scene initialization prompts. Secondly, the hierarchical prompt generator, based on scene prompt templates, decomposes the reasoning steps into continuous thought chain nodes, generating thought chain guidance prompts. Furthermore, the hierarchical suggestion generator collects historical optimization results, evaluation feedback, and the best expression of the current search to form integrated feedback suggestions. and guided by mind chain prompts In conjunction with this, the prompts are dynamically updated to guide subsequent optimizations;

[0008] The chain-based reasoning engine first uses the symbolic feature space Construct the initial symbolic expression population. And use a genetic evolution algorithm to perform several rounds of search iterations to optimize the current time step. symbolic expression population Secondly, the chain-based reasoning engine utilizes the thought chain in the hierarchical hint generator to guide hints. Integration of feedback prompts Input the current context information into the large language model, and let the large language model provide optimization suggestions based on its symbolic reasoning and scene patterns. During this process, the mind chain guides and prompts. Used to progressively unfold the reasoning process and provide feedback and integrated prompts. Continuously deepen the understanding of logical expressions and provide optimization suggestions;

[0009] After receiving the inference results from the current large language model, the symbolic expression optimizer uses a mapping function. Transform the reasoning results into a symbolic feature space. Configured executable optimization operations This is added to the symbolic expression population of the symbolic search; the optimization process adopts an iterative strategy, continuously evaluating and improving the quality of the expression until the predetermined optimization objective or convergence condition is reached.

[0010] The scene information includes feature information such as scene entity categories, location information, and category counts, as well as a set of symbolic operators specified by the user.

[0011] The entire prompt generation process of the hierarchical prompt generator is represented as follows:

[0012] ;

[0013] in, This represents the generated prompt sequence; Represents a hierarchical suggestion generator; Representative scenario prompt template; This represents the scene information and context information obtained.

[0014] The reasoning process of the chain-based reasoning engine is represented as follows:

[0015] ;

[0016] in, This represents the reasoning results, including optimization suggestions and possible directions for improvement; Represents a chain-based reasoning engine; Mind chain guidance prompts representing the input chain reasoning engine Integration of feedback prompts ; The best individual in the population of expressions.

[0017] In the chain-based reasoning engine, the constructed symbolic expression population contains several individual symbolic expressions used to match scene patterns; firstly, the chain-based reasoning engine will, based on the current time... symbolic expression population Perform several rounds of logical search and construct the population after the search. Then the searched population The corresponding fitness assessment results are fed into a hierarchical prompt generator, where they are combined with a pre-set feedback integration prompt template to construct feedback integration prompts. The process is represented as:

[0018] ;

[0019] in, Representative scenario prompt template, Represents the population after the search The evaluation results;

[0020] Guide the thought process Integration tips with feedback The data are organically combined and fed into a large language model to obtain new reasoning suggestions that have undergone symbolic reasoning and common sense judgment by the large language model.

[0021] Preferably, the symbolic expression optimizer first extracts usable inference results from the inference suggestions of the large language model. Secondly, the symbolic expression optimizer combines the symbolic feature space of the current scene. The reasoning results This process of transforming optimizations into concrete, executable operations can be represented as follows: After evaluating the fitness of these executable optimization operations, they are updated into the population at the next time step to obtain the updated population. This process is repeated until the termination condition is met. Then, the symbolic expression with the highest fitness is obtained from the latest symbolic expression population and used as the system output. This output is a symbolic expression that conforms to the scenario pattern constructed by the user.

[0022] Preferably, the termination condition is the number of iterations or a fitness target.

[0023] After adopting the above technical solution, the present invention has the following technical effects:

[0024] ① This invention focuses on the optimization problem of symbolic logic reasoning, and proposes a novel symbolic logic reasoning optimization paradigm, which successfully achieves deep integration of large language models and symbolic reasoning systems, and improves the matching degree of logical expressions to visual scenes;

[0025] ②Through the spatial design of layered prompts, the system can effectively guide the large language model to engage in structured thinking, enabling it to accurately understand the scene features provided by the user;

[0026] ③ The reasoning mechanism based on chain thinking ensures the coherence and reliability of the reasoning process;

[0027] ④ The innovative symbol optimization strategy enables a seamless transition from language understanding to logic optimization;

[0028] ⑤ It can also generate reasonable optimization suggestions through the reasoning ability of large language models, and effectively transform these suggestions into specific optimization operations;

[0029] ⑥ The modular design of the system not only ensures the clarity of the processing flow, but also facilitates subsequent functional expansion and performance optimization. Attached Figure Description

[0030] Figure 1 This is a flowchart of a specific embodiment of the present invention. Detailed Implementation

[0031] To further explain the technical solution of the present invention, the present invention will be described in detail below through specific embodiments.

[0032] refer to Figure 1 As shown, this invention discloses a symbolic logic reasoning optimization system based on a large language model. This system achieves end-to-end processing from scene input to optimization result output through the collaborative work of three core modules: a hierarchical prompt generator, a chained reasoning engine, and a symbolic expression optimizer. The system first constructs an initial symbolic feature space based on the scene information input by the user. This provides the basic operating environment for the subsequent inference and optimization process. The scene information includes feature information such as scene entity categories, location information, and category counts, as well as a set of symbolic operators specified by the user. The work of the three core modules is as follows:

[0033] (1) Hierarchical prompt generator

[0034] This module employs a carefully designed multi-layered prompting architecture, including scene initialization prompts. Mind chain guidance prompts Integration of feedback prompts Specifically:

[0035] First, the hierarchical hint generator will use the symbol feature space. The information is combined with the preset scene prompt template to generate scene initialization prompts. This is used to activate the prior knowledge of a large language model for a specific scenario;

[0036] Secondly, the hierarchical prompt generator, based on scene prompt templates, decomposes reasoning steps into continuous thought chain nodes, generating thought chain guidance prompts. This is used to help large language models break down the big problem of merging scene analysis and symbolic reasoning into multiple smaller problems;

[0037] Furthermore, the hierarchical suggestion generator collects historical optimization results, evaluation feedback, and the best expression of the current search to form integrated feedback suggestions. and guided by mind chain prompts In conjunction with this, the prompts are dynamically updated to guide subsequent optimizations.

[0038] The entire hint generation process of the hierarchical hint generator can be represented as:

[0039] ;

[0040] in, This represents the generated prompt sequence; Represents a hierarchical suggestion generator; Representative scenario prompt template; This represents the acquired scene and context information. The generated prompt sequence. It was then used to guide the reasoning process of large language models.

[0041] (2) Chain-based reasoning engine

[0042] This engine implements an innovative reasoning mechanism based on a large language model; specifically:

[0043] First, the chain-based reasoning engine uses the symbolic feature space. Construct the initial symbolic expression population. And use a genetic evolution algorithm to perform several rounds of search iterations to optimize the current time step. symbolic expression population ;

[0044] Secondly, the chain-based reasoning engine utilizes the thought chain in the hierarchical hint generator to guide hints. Integration of feedback prompts Input the current context information into the large language model, and let the large language model provide optimization suggestions based on its symbolic reasoning and scene patterns. During this process, the mind chain guides and prompts. Used to progressively unfold the reasoning process and provide feedback and integrated prompts. Continuously deepen the understanding of logical expressions and provide optimization suggestions.

[0045] The reasoning process of a chain-based reasoning engine can be represented as:

[0046] ;

[0047] in, This represents the reasoning results, including optimization suggestions and possible directions for improvement; Represents a chain-based reasoning engine; Mind chain guidance prompts representing the input chain reasoning engine Integration of feedback prompts ; This represents the optimal individual in the population of expressions. The engine employs a fitness-based selection mechanism to ensure the degree to which the inference result matches the current scenario.

[0048] (3) Symbolic expression optimizer

[0049] This module is responsible for transforming the inference results of the large language model into specific optimization operations; specifically:

[0050] After receiving the inference results from the current large language model, the mapping function is used. Transform the reasoning results into a symbolic feature space. Configured executable optimization operations This is then added to the symbolic expression population for symbolic search. The optimization process employs an iterative strategy, continuously evaluating and improving the quality of the expression until the predetermined optimization objective or convergence condition is met.

[0051] Through the above-described scheme, this invention focuses on the optimization problem of symbolic logic reasoning, proposing a novel paradigm for symbolic logic reasoning optimization. It successfully achieves deep integration of a large language model and a symbolic reasoning system (this is the core innovation of this invention), improving the matching degree of logical expressions to visual scenes. Through a hierarchical prompting spatial design, the system effectively guides the large language model to engage in structured thinking, enabling it to accurately understand the scene features provided by the user. A chain-based reasoning mechanism ensures the coherence and reliability of the reasoning process. The innovative symbolic optimization strategy achieves a seamless transition from language understanding to logical optimization. Furthermore, it can generate reasonable optimization suggestions through the reasoning capabilities of the large language model and effectively translate these suggestions into specific optimization operations. The modular design of the system not only ensures the clarity of the processing flow but also facilitates subsequent functional expansion and performance optimization.

[0052] In terms of technical effectiveness, this invention demonstrates significant advantages: Regarding reasoning efficiency, the system, guided by a large language model, improves search speed by 60% in visual event detection tasks, and the average pattern matching index improves by 12%, significantly enhancing optimization efficiency. Simultaneously, the interpretability of the logical expressions generated by the system is significantly improved while maintaining high logical rigor. More importantly, the system exhibits excellent practical value; through collaboration with a large language model, it can understand and optimize a wider range of logical expressions. Furthermore, the reasoning process and optimization suggestions provided by the system are highly interpretable, facilitating user understanding and verification.

[0053] These characteristics make this invention not only technically innovative, but also demonstrate significant advantages and value in practical applications, providing a new paradigm and approach for the development of symbolic logic reasoning technology.

[0054] The following illustrates specific embodiments of the present invention.

[0055] In the aforementioned chain-based reasoning engine, the constructed symbolic expression population contains several individual symbolic expressions used to match scene patterns; firstly, the chain-based reasoning engine will, based on the current time... symbolic expression population Perform several rounds of logical search and construct the population after the search. Then the searched population The corresponding fitness evaluation results are fed into a hierarchical cue generator, where they are combined with a pre-defined feedback integration cue template to construct feedback integration cue. This process can be represented as:

[0056] ;

[0057] in, Representative scenario prompt template, Represents the population after the search The evaluation results;

[0058] Guide the thought process Integration tips with feedback The data are organically combined and fed into a large language model to obtain new reasoning suggestions that have undergone symbolic reasoning and common sense judgment by the large language model.

[0059] Furthermore, since the reasoning suggestions provided by the large language model are text composed of natural language, they cannot be directly incorporated into the symbolic expression population to participate in the next round of iteration. Therefore, this invention uses a symbolic expression optimizer:

[0060] First, extract usable reasoning results from the reasoning suggestions of the large language model. ;

[0061] Secondly, combining the symbolic feature space of the current scene The reasoning results Transforming this into concrete, executable optimization operations (individual expressions) can be represented as follows: After evaluating the fitness of these executable optimization operations, they are updated into the population at the next time step to obtain the updated population. This process is repeated until a termination condition is met (which can be the number of iterations or a fitness target). Then, the symbolic expression with the highest fitness is obtained from the latest symbolic expression population and used as the system output. This output is a symbolic expression that conforms to the scenario pattern constructed by the user.

[0062] The above embodiments and figures are not intended to limit the product form and style of the present invention. Any appropriate changes or modifications made by those skilled in the art should be considered as not departing from the patent scope of the present invention.

Claims

1. A symbolic logic reasoning optimization system based on a large language model, characterized in that: It includes three core modules: a hierarchical prompt generator, a chained reasoning engine, and a symbolic expression optimizer. It realizes an end-to-end processing process from the input of scene information to the output of optimized reasoning results from a large language model. The system outputs symbolic expressions that conform to scene patterns and are constructed based on the scene information provided by the user. The system improves the search speed in visual event detection tasks through the guidance of the large language model, and through collaboration with the large language model, the system can understand and optimize a wider range of logical expressions. The system first constructs an initial symbolic feature space based on the scene information input by the user. The scene information includes feature information such as scene entity categories, location information, and category counts, as well as a set of symbolic operators specified by the user; the operation of the three core modules is as follows: The multi-level prompting architecture of the layered prompt generator includes scene initialization prompts. Mind chain guidance prompts Integration of feedback prompts First, the hierarchical hint generator will use the symbol feature space. The information is combined with the preset scene prompt template to generate scene initialization prompts. This is used to activate the prior knowledge of the scene in the large language model; Secondly, the hierarchical prompt generator, based on scene prompt templates, decomposes the reasoning steps into continuous thought chain nodes, generating thought chain guidance prompts. This is used to help large language models break down the large problem of merging scene analysis and symbolic reasoning into multiple smaller problems; furthermore, the hierarchical suggestion generator collects historical optimization results, evaluation feedback, and the best expression of the current search to form feedback integration suggestions. and guided by mind chain prompts In conjunction with this, dynamic updates to the prompts will guide subsequent optimizations. The chain-based reasoning engine first uses the symbolic feature space Construct the initial symbolic expression population. And use a genetic evolution algorithm to perform several rounds of search iterations to optimize the current time step. symbolic expression population ; Secondly, the chain-based reasoning engine utilizes the thought chain in the hierarchical hint generator to guide hints. Integration of feedback prompts Input the current context information into the large language model, and let the large language model provide optimization suggestions based on its symbolic reasoning and scene patterns. During this process, the mind chain guides and prompts. Used to progressively unfold the reasoning process and provide feedback and integrated prompts. Continuously deepen the understanding of logical expressions and provide optimization suggestions; the reasoning process of the chain-based reasoning engine is represented as follows: ； in, This represents the reasoning results, including optimization suggestions and possible directions for improvement; Represents a chain-based reasoning engine; Mind chain guidance prompts representing the input chain reasoning engine Integration of feedback prompts ; The best individual in the population representing the expression; The chain-based inference engine employs a fitness-based filtering mechanism to ensure the degree of matching between the inference results and the current scenario. Within the chain-based inference engine, a constructed population of symbolic expressions contains several individual symbolic expressions used to match scenario patterns. First, the chain-based inference engine will, based on the current time... symbolic expression population Perform several rounds of logical search and construct the population after the search. Then the searched population The corresponding fitness assessment results are fed into a hierarchical prompt generator, where they are combined with a pre-set feedback integration prompt template to construct feedback integration prompts. The process is represented as: ； in, Representative scenario prompt template, Represents the population after the search The evaluation results; Guide the thought process Integration tips with feedback The data are organically combined and fed into a large language model to obtain new reasoning suggestions that have undergone symbolic reasoning and common sense judgment by the large language model. These new reasoning suggestions are text composed of natural language. The symbolic expression optimizer is responsible for transforming the inference results of the large language model into executable optimization operations, achieving deep integration between the large language model and the symbolic logic inference optimization system. Specifically, after receiving the inference results of the current large language model, it uses a mapping function... Transform the reasoning results into a symbolic feature space. Configured executable optimization operations After evaluating the fitness of these executable optimization operations, they are updated into the population at the next time step to obtain the updated population. The expression is added to the symbolic expression population for symbolic search. The optimization process adopts an iterative strategy, continuously evaluating and improving the quality of the expression until the predetermined optimization objective or convergence condition is reached. The symbolic expression with the highest fitness is obtained from the latest symbolic expression population as the output of the system. The termination condition of the iterative strategy is the number of iterations or the fitness target.

2. The symbolic logic reasoning optimization system based on a large language model as described in claim 1, characterized in that: The entire prompt generation process of the hierarchical prompt generator is represented as follows: ； wherein, representing a generated shot sequence; representing a hierarchical shot generator; representing a scene shot template; representing acquired scene information and context information.

Images