Intelligent reasoning method and system based on large language model

Abstract

The invention discloses an intelligent reasoning method and system based on a large language model, and belongs to the technical field of intelligent reasoning, and the method specifically comprises the steps: obtaining input data, carrying out the preprocessing, recognizing a current situation and a reasoning target on the basis of the input data, and carrying out the recognition of the current situation and the reasoning target through the combination of a deep neural network and the large language model, and employing a recursive reasoning mechanism. Reasoning analysis is carried out in stages, and a final reasoning result is output; by means of the method, the accuracy and reliability of the reasoning result are improved, the method is particularly excellent in performance under complex application scenes and complex tasks, adjustment can be conducted according to specific requirements of different users, and the method has the high individuation capacity.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent reasoning technology, and specifically relates to an intelligent reasoning method and system based on a large language model. Background Art

[0002] With the rapid development of artificial intelligence technology, large language models (such as GPT, BERT, etc.) have made significant progress in the field of natural language processing (NLP). These models can understand and generate human language through large-scale pre-training and fine-tuning, and are widely used in tasks such as text generation, machine translation, and question-answering systems. However, traditional language models still have limitations when dealing with complex reasoning tasks, especially in scenarios that require logical reasoning, multi-step reasoning, and cross-domain knowledge fusion.

[0003] Existing reasoning methods usually rely on rule engines, knowledge graphs, or symbolic logic. Although these methods perform well in certain specific areas, they lack flexibility and generalization capabilities. The emergence of large language models provides new possibilities for solving these problems. By combining deep learning and natural language understanding, large language models can perform reasoning without explicit rules, but their reasoning capabilities still need to be further optimized and expanded.

[0004] For example, a Chinese patent with authorization announcement number CN116992966B discloses a method and computing device for an artificial intelligence model reasoning platform. The method includes: obtaining first input data for a first reasoning task and executing the first reasoning task; obtaining second input data for a second reasoning task and executing the second reasoning task; if the second input data is the same as the first input data, and the second reasoning path is the same as the first reasoning path, the second reasoning task reuses the output data of the first reasoning path. According to the technical solution of the invention, the amount of algorithm calculation can be effectively reduced, the computing power of the chip can be reasonably utilized, and resource waste can be avoided.

[0005] For example, a Chinese patent with authorization announcement number CN118485150B discloses a method for generating reasoning information based on artificial intelligence, including: obtaining the user's input text, splitting it to obtain the knowledge graph of the input text; obtaining several main nodes and several description nodes; obtaining the description exclusivity of each description node for each main node; obtaining the main description degree of different jump neighborhoods of each main node, and then obtaining the effective neighborhood range of each main node; obtaining the description accuracy of each main node; obtaining several important semantic windows based on the description accuracy and effective neighborhood range of each main node; updating the knowledge graph based on the important semantic windows, and completing text reasoning through local graph encoding. The invention aims to solve the problem of reasoning information bias caused by insufficient mining of connections between unconnected nodes in the knowledge graph.

[0006] The defects of the above patents: lack of personalization and adaptability, and problems with real-time and dynamics. Summary of the invention

[0007] In view of the shortcomings of the existing technology, the present invention proposes an intelligent reasoning method and system based on a large language model. By integrating deep learning technology with a large language model, a multi-level, recursively optimized intelligent reasoning system is constructed to solve the problems faced in reasoning and analysis tasks in complex scenarios. By combining historical data, external knowledge bases and contextual information, the present invention can improve the accuracy, interpretability and adaptability of reasoning results.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] The intelligent reasoning method based on the large language model includes the following specific steps:

[0010] Get input data and perform preprocessing;

[0011] Based on the input data, identify the current situation and reasoning goals;

[0012] By combining deep neural networks with large language models and adopting a recursive reasoning mechanism, reasoning analysis is performed in stages to output the final reasoning results.

[0013] Specifically, based on the input data, identifying the current situation and reasoning target includes:

[0014] Analyze the preprocessed input data and perform scenario modeling based on the analysis results;

[0015] Based on situation modeling, combined with preprocessed input data, the goal of the current task is derived;

[0016] Break down the task objectives into sub-goals and adjust the task sub-goals in real time according to changes in the situation;

[0017] The relationship between task objectives and situations is modeled through multi-level reasoning networks and graph neural networks, and the constraint relationship between situational factors and task objectives is established. The constraint conditions of task objectives are perceived and adjusted in real time according to situational changes.

[0018] Based on historical data and external knowledge base, the current task objectives are further refined and predicted.

[0019] Specifically, the context modeling includes: context modeling of text data, context modeling of speech data, and context modeling of image data.

[0020] Specifically, the method combines a deep neural network with a large language model, adopts a recursive reasoning mechanism, performs reasoning analysis in stages, and outputs the final reasoning results, including:

[0021] Based on the recognition of context and task objectives, deep neural networks and large language models are used to analyze existing data and generate preliminary reasoning results;

[0022] Analyze the preliminary reasoning results, combine historical data, external knowledge base and context information, and optimize the preliminary reasoning results;

[0023] Perform multi-path reasoning through a recursive mechanism, automatically select the appropriate reasoning path according to different situations, perform local optimization on each path, and evaluate the reasoning results of each path after optimization;

[0024] The evaluation results of each path are sorted, and the top N paths in the evaluation results are selected for aggregation to obtain the final inference result.

[0025] Specifically, the input data includes: text, voice, image, sensor data, structured table data or log file.

[0026] Specifically, the preprocessing includes: data cleaning, word segmentation, stop word removal and stem extraction for text data, and vectorization of text using a pre-trained large language model;

[0027] Voice data, using speech recognition systems to convert speech into text;

[0028] Image data, denoising and enhancement, using convolutional neural networks to extract high-level feature vectors of images;

[0029] Fusion of text vectors, speech vectors and image feature vectors.

[0030] An intelligent reasoning system based on a large language model, used to implement the intelligent reasoning method based on a large language model, comprising: a data processing module, a situation and target analysis module and a reasoning module;

[0031] The data processing module is used to obtain input data and perform preprocessing;

[0032] The context and goal analysis module is used to identify the current context and reasoning goal based on the input data;

[0033] The reasoning module is used to combine a deep neural network with a large language model, adopt a recursive reasoning mechanism, perform reasoning analysis in stages, and output the final reasoning result.

[0034] Specifically, the situation and target analysis module includes: a situation analysis unit and a target analysis unit;

[0035] The situation analysis unit is used to analyze the preprocessed input data and perform situation modeling according to the analysis results;

[0036] The target analysis unit is used to derive the target of the current task, break down the target of the task into sub-targets, adjust the task sub-targets in real time according to changes in the situation, establish a constraint relationship between situational factors and task targets, and adjust the task targets in real time based on historical data and an external knowledge base.

[0037] Specifically, the reasoning module includes: a preliminary result unit and a final reasoning result unit;

[0038] The preliminary result unit is used to analyze the existing data using a deep neural network and a large language model based on the recognition of the context and the task goal, and generate preliminary reasoning results;

[0039] The final reasoning result unit is used to optimize the preliminary reasoning result and aggregate the paths using a recursive multi-path mechanism to obtain the final reasoning result.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] 1. The present invention proposes an intelligent reasoning method based on a large language model, which can be adjusted according to the specific needs of different users and has a high degree of personalization.

[0042] 2. The present invention proposes an intelligent reasoning method based on a large language model, which can process and update the reasoning results in real time, and can dynamically adjust the reasoning path according to new data input (such as real-time examination results, patient feedback, etc.) to ensure that the reasoning results are in line with the latest reality.

[0043] 3. The present invention proposes an intelligent reasoning method based on a large language model, which can integrate knowledge from different fields, support cross-domain reasoning tasks, and realize multi-step reasoning and logical inference in complex tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A flow chart of the intelligent reasoning method based on a large language model provided by the present invention;

[0045] Figure 2 This is an architecture diagram of the intelligent reasoning system based on a large language model provided by the present invention. DETAILED DESCRIPTION

[0046] The present application is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present application, but do not limit the present application in any form. It should be noted that, for those of ordinary skill in the art, several variations and improvements can also be made without departing from the concept of the present application. These all belong to the protection scope of the present application.

[0047] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0048] It should be noted that, if there is no conflict, the various features in the embodiments of the present application can be combined with each other, all within the scope of protection of the present application. In addition, although the functional module division is performed in the device schematic diagram and the logical order is shown in the flow chart, in some cases, the steps shown or described can be performed in a sequence different from the module division in the device or the flow chart. In addition, the words "first", "second", "third" and the like used in the present application do not limit the data and the execution order, but only distinguish the same items or similar items with substantially the same functions and effects.

[0049] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used in this specification and in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The term "and / or" used in this specification includes any and all combinations of one or more of the related listed items.

[0050] Example 1

[0051] See also Figure 1 , an embodiment provided by the present invention: an intelligent reasoning method based on a large language model, comprising the following specific steps:

[0052] Step S1: Obtain input data and perform preprocessing;

[0053] The input data in step S1 includes but is not limited to text, voice, image, sensor data, structured table data or log file, etc. In this step, the data source will be automatically marked, and the corresponding preprocessing process will be performed according to the type of data source. For example: for text data, the system uses natural language processing tools (such as word segmentation, part-of-speech tagging, etc.) to perform preliminary text cleaning; for voice data, a voice recognition system is used to convert voice into text; for image data, image recognition technology is used to extract key features or labels in the image; at the same time, at this stage, data that does not meet the quality standards (such as parts with excessive noise or lack of effective information) are preliminarily screened to eliminate unnecessary noise and erroneous data to ensure the accuracy and efficiency of subsequent processing;

[0054] Preprocessing also includes multimodal fusion and standardization. Text data standardization: by means of word segmentation, stop word removal, stem extraction, etc., the original text is converted into a unified semantic representation. Furthermore, the text is vectorized with the help of a pre-trained large language model so as to capture the deep semantic relationship between words. Image data standardization: when preprocessing images, image enhancement techniques (such as rotation, scaling, flipping, etc.) are used to increase data diversity and reduce data deviation. At the same time, convolutional neural networks (CNNs) are used to extract high-level feature vectors of images, and these features are connected with other modal data. Speech data processing: speech data is converted into text through automatic speech recognition (ASR) technology. This is followed by the use of natural language processing technology to analyze the text content. If the voice data comes from a dialogue system, sentiment analysis and tone recognition technology will be used to extract the emotion or context information of the voice; Structured data processing: First, the table data is standardized, missing values ​​are filled, outliers are removed, and normalization and standardization are performed. Then, the relationship between structured data is modeled using graph neural networks (GNNs) to form feature vectors that are fused with unstructured data; Unstructured data processing: Unstructured data (such as images, texts) are feature extracted through convolutional neural networks, recursive neural networks, etc. These features will be fused with the features of structured data in the same feature space;

[0055] Finally, multimodal fusion is performed. The goal of the fusion is to construct a multi-dimensional feature space so that data from different modalities can be effectively integrated under the same reasoning framework.

[0056] Step S2: Based on the input data, identify the current situation and reasoning goal;

[0057] The specific steps of step S2 are:

[0058] Step S201: Analyze the pre-processed input data and perform situation modeling based on the analysis results;

[0059] In this embodiment, context modeling includes three categories: 1) context modeling of text data: for input text data, semantic understanding is performed through large language models (such as BERT, GPT, etc.). These models will not only identify the relationship between keywords and sentences, but also analyze the context and potential emotions behind the text. For example, in a medical scenario, the system will identify the symptoms, drugs and disease names mentioned by the patient, and understand the patient's condition background in combination with the context to form a complete context description; 2) context modeling of voice data: for voice data, first use automatic speech recognition (ASR) technology to convert the voice into text, and then use sentiment analysis technology to identify tone, emotion and implied meaning to ensure that the potential emotions and intentions in the conversation are accurately captured. For example, if the user's tone appears anxious, the system will adjust the reasoning path according to this emotional change; 3) context modeling of image data: for image data, especially medical images, satellite images, etc., deep analysis is performed through convolutional neural networks (CNN) to extract key features in the image (such as lesion area, object location, etc.). Further, the system combines the image content with the external knowledge base to generate a context description corresponding to the image, and combines it with the text data to form a multimodal context understanding.

[0060] Step S202: Based on the situation modeling and combined with the pre-processed input data, derive the goal of the current task;

[0061] For example, in an enterprise decision support scenario, the task goal may be "increase product sales" or "optimize inventory management", while in a medical scenario, the task goal may be "diagnose a disease" or "determine a treatment plan". Natural language processing (NLP) technology is used to identify the user's intention and abstract the goal.

[0062] Step S203: breaking down the task goal into several sub-goals, and adjusting the task sub-goals in real time according to changes in the situation;

[0063] For example, in a medical task, if the task goal is "diagnose a disease", it can be further refined into sub-goals such as "analyze patient symptoms", "retrieve related cases", and "recommend possible treatment plans". When the situation changes, such as: the patient mentions a related case, the task goal is refined into sub-goals such as "analyze patient symptoms" and "recommend possible treatment plans".

[0064] Step S204: Model the relationship between the task goal and the context through a multi-level reasoning network and a graph neural network (GNN), establish a constraint relationship between the context factors and the task goal, and perceive and adjust the constraint conditions of the task goal in real time according to the context changes;

[0065] In this embodiment, in financial decision-making, the task goal may be "predicting stock market trends", and the situational factors may include "economic data", "market sentiment", etc. These different dimensions of information are connected together through the graph neural network to establish a constraint relationship between the situational factors and the task goal. If some situational factors change (such as economic crisis, sudden changes in customer demand, etc.), these changes are automatically identified, and the task goal is optimized or adjusted to ensure the effectiveness and rationality of the final decision;

[0066] Step S205: Based on historical data and external knowledge base, further refine and predict the current task objectives.

[0067] In this embodiment, historical data analysis: based on similar situations and task execution situations in historical data, pattern recognition and trend deduction are performed. For example, in customer service scenarios, historical conversation records are analyzed to infer the needs of current users and identify goals related to user needs. By learning common problems, solutions and customer feedback in historical data, potential task goals in the current situation can be quickly located. External knowledge base docking and reasoning: through connection with external knowledge bases (such as professional knowledge bases in the fields of medicine, law, finance, etc.), domain knowledge can be referenced when identifying the current task goals to ensure the comprehensiveness and accuracy of the goals. For example, in medical reasoning tasks, external knowledge bases can provide the latest disease diagnosis and treatment guidelines, drug treatment plans and other information to further refine task goals and ensure that task goals meet domain requirements.

[0068] Step S3: By combining the deep neural network with the large language model and adopting the recursive reasoning mechanism, the reasoning analysis is performed in stages and the final reasoning result is output.

[0069] The specific steps of step S3 are:

[0070] Step S301: Based on the recognition of the context and task objectives, the deep neural network and the large language model are used to analyze the existing data (such as the literature database, etc.) to generate preliminary reasoning results;

[0071] In this embodiment, this step is the second stage of recursive reasoning (the first stage is data collection and data processing). The reasoning results output in this stage are only preliminary diagnoses or suggestions, but provide important basis for subsequent reasoning;

[0072] Step S302: Analyze the preliminary reasoning results, combine historical data, external knowledge base and context information, and optimize the preliminary reasoning results;

[0073] In this embodiment, historical case review: data source, historical case library, patient health records, known disease patterns, etc.; analysis target, identifying historical cases similar to the current patient's condition, and helping to optimize the preliminary reasoning results by reviewing the treatment results, disease progression, etc. of these cases; application method: for example, for early diagnosis of a disease, if the preliminary reasoning results suggest a certain disease, cases with similar symptoms and course of disease can be retrieved from historical cases, and the symptom manifestations, treatment methods and treatment effects in these cases can be analyzed to help exclude mismatched diseases or adjust the diagnosis; course of disease and treatment effect data: data source, patient treatment history, medication records, treatment response, etc.; analysis target, judging the patient's treatment response and drug efficacy, and correcting the preliminary reasoning results in combination with historical data. For example, the reasoning of the treatment effect of a certain drug can be corrected by the patient's response to the drug in historical cases (improvement, ineffectiveness or side effects, etc.);

[0074] Integrate external knowledge bases, including professional knowledge maps in the medical field, academic literature, research databases, etc. These data sources can provide the latest medical research results, drug information, disease classification and other information for preliminary reasoning;

[0075] Dynamic integration of contextual information. Contextual information refers to information such as the patient's current living environment, health status, clinical symptoms and their changing process. These factors have an important impact on the inference results and treatment decisions of the disease, including: real-time health status analysis, data source, real-time health data of patients, such as body temperature, blood pressure, laboratory test results, imaging data, etc.; analysis goals: real-time monitoring of changes in the patient's health status, and comparing it with the preliminary inference results, and making dynamic adjustments. For example, if the patient's temperature rises and is accompanied by new symptoms, the inference results can be adjusted to increase the probability of possible infectious diseases (such as influenza, pneumonia, etc.); dynamic changes in clinical symptoms, data source, symptoms Change records, patient-reported self-perception, etc.; analysis goals, by monitoring the changing trends of symptoms (such as worsening or relief of symptoms) to adjust the reasoning path, and by dynamically updating the symptom description, make the reasoning results more consistent with the current clinical manifestations of the patient; application methods, for example, if the patient's symptoms change during the treatment process (such as worsening headache, increased cough), re-evaluate the possibility of the disease according to the changes in symptoms, and make corresponding adjustments in the reasoning path; consideration of environment and living habits, combined with the patient's living habits and environmental factors, optimize the preliminary reasoning results. For example, if the patient has recently been in contact with the epidemic area, or has a history of exposure to certain occupational diseases, it will increase the probability of diseases related to these risk factors.

[0076] Exemplarily, medication recommendations are tailored based on the patient's specific history, medication side effects, and other medical literature;

[0077] Step S303: Perform multi-path reasoning through a recursive mechanism, that is, explore multiple possible reasoning paths, automatically select the appropriate reasoning path according to different situations, perform local optimization on each path, and evaluate the reasoning results of each path after optimization;

[0078] In this embodiment, the generated path can be inferred by using a deep neural network model and a large language model respectively. If the results of the two models are consistent, the credibility of the results is enhanced; if there are differences, the system will further analyze and adjust the final results according to the context information and historical data. After local optimization of each path, dynamic feedback will be provided to adjust the inference results according to real-time input information. For example, if the user reports that his or her symptoms have worsened or new medical data is input, the inference results will be dynamically updated and recursive optimization will continue;

[0079] Step S304: Sort the evaluation results of each path, select the top N paths in the evaluation results for aggregation, and obtain the final inference result.

[0080] For example, in a medical scenario, the system may aggregate multiple possible drug recommendations and output a comprehensive recommendation result.

[0081] Example 2

[0082] See also Figure 2 ,Another embodiment provided by the present invention: An intelligent reasoning system based on a large language model, comprising: a data processing module, a context and target analysis module and a reasoning module;

[0083] The data processing module is used to obtain input data and perform preprocessing;

[0084] The context and goal analysis module is used to identify the current context and reasoning goal based on the input data;

[0085] The reasoning module is used to combine a deep neural network with a large language model, adopt a recursive reasoning mechanism, perform reasoning analysis in stages, and output the final reasoning result.

[0086] The situation and goal analysis module includes: situation analysis unit and goal analysis unit;

[0087] The situation analysis unit is used to analyze the preprocessed input data and perform situation modeling according to the analysis results;

[0088] The target analysis unit is used to derive the target of the current task, break down the target of the task into sub-targets, adjust the task sub-targets in real time according to changes in the situation, establish a constraint relationship between situational factors and task targets, and adjust the task targets in real time based on historical data and an external knowledge base.

[0089] The reasoning module includes: a preliminary result unit and a final reasoning result unit;

[0090] The preliminary result unit is used to analyze the existing data using a deep neural network and a large language model based on the recognition of the context and the task goal, and generate preliminary reasoning results;

[0091] The final reasoning result unit is used to optimize the preliminary reasoning result and aggregate the paths using a recursive multi-path mechanism to obtain the final reasoning result.

[0092] In addition, the parts of the above-mentioned technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive redundancy.

[0093] The specific implementation modes as described above further describe the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation mode of the present invention and is not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An intelligent reasoning method based on a large language model, characterized in that: include: Get input data and perform preprocessing; Based on the input data, identify the current situation and reasoning goals; By combining deep neural networks with large language models and adopting a recursive reasoning mechanism, reasoning analysis is performed in stages to output the final reasoning results.

2. The intelligent reasoning method based on a large language model as claimed in claim 1, characterized in that: Based on the input data, the current situation and reasoning goal are identified, including: Analyze the preprocessed input data and perform scenario modeling based on the analysis results; Based on situation modeling, combined with preprocessed input data, the goal of the current task is derived; Break down the task objectives into sub-goals and adjust the task sub-goals in real time according to changes in the situation; The relationship between task objectives and situations is modeled through multi-level reasoning networks and graph neural networks, and the constraint relationship between situational factors and task objectives is established. The constraint conditions of task objectives are perceived and adjusted in real time according to situational changes. Based on historical data and external knowledge base, the current task objectives are further refined and predicted.

3. The intelligent reasoning method based on a large language model as claimed in claim 2, characterized in that: The context modeling includes: context modeling of text data, context modeling of voice data and context modeling of image data.

4. The intelligent reasoning method based on a large language model as claimed in claim 1, characterized in that: The method combines a deep neural network with a large language model, adopts a recursive reasoning mechanism, performs reasoning analysis in stages, and outputs the final reasoning results, including: Based on the recognition of context and task objectives, deep neural networks and large language models are used to analyze existing data and generate preliminary reasoning results; Analyze the preliminary reasoning results, combine historical data, external knowledge base and context information, and optimize the preliminary reasoning results; Perform multi-path reasoning through a recursive mechanism, automatically select the appropriate reasoning path according to different situations, perform local optimization on each path, and evaluate the reasoning results of each path after optimization; The evaluation results of each path are sorted, and the top N paths in the evaluation results are selected for aggregation to obtain the final inference result.

5. The intelligent reasoning method based on a large language model as claimed in claim 1, characterized in that: The input data includes: text, voice, image, sensor data, structured table data or log file.

6. The intelligent reasoning method based on a large language model as claimed in claim 1, characterized in that: The preprocessing includes: data cleaning, word segmentation, stop word removal and stem extraction for text data, and vectorization of text using a pre-trained large language model; Voice data, using speech recognition systems to convert speech into text; Image data, denoising and enhancement, using convolutional neural networks to extract high-level feature vectors of images; Fusion of text vectors, speech vectors and image feature vectors.

7. An intelligent reasoning system based on a large language model, used to implement the intelligent reasoning method based on a large language model as described in any one of claims 1 to 6, characterized in that: include: Data processing module, situation and goal analysis module and reasoning module; The data processing module is used to obtain input data and perform preprocessing; The context and goal analysis module is used to identify the current context and reasoning goal based on the input data; The reasoning module is used to combine a deep neural network with a large language model, adopt a recursive reasoning mechanism, perform reasoning analysis in stages, and output the final reasoning result.

8. The intelligent reasoning system based on a large language model as claimed in claim 7, characterized in that: The situation and target analysis module includes: a situation analysis unit and a target analysis unit; The situation analysis unit is used to analyze the preprocessed input data and perform situation modeling according to the analysis results; The target analysis unit is used to derive the target of the current task, break down the target of the task into sub-targets, adjust the task sub-targets in real time according to changes in the situation, establish a constraint relationship between situational factors and task targets, and adjust the task targets in real time based on historical data and an external knowledge base.

9. The intelligent reasoning system based on a large language model as claimed in claim 8, characterized in that: The reasoning module includes: a preliminary result unit and a final reasoning result unit; The preliminary result unit is used to analyze the existing data using a deep neural network and a large language model based on the recognition of the context and the task goal, and generate preliminary reasoning results; The final reasoning result unit is used to optimize the preliminary reasoning result and aggregate the paths using a recursive multi-path mechanism to obtain the final reasoning result.

Images