Generative AI text reasoning and characterization method based on loyalty perception mechanism

Abstract

The invention provides a generative AI text reasoning and characterization method based on a loyalty perception mechanism, and relates to the field of artificial intelligence, and the method comprises the steps: constructing a complete thinking chain, and guiding the generative analysis content; cutting off the complete thinking chain, collecting a plurality of AI analysis contents generated under the guidance of the cut thinking chain, and measuring the loyalty; calculating a sentence-level comprehensive loyalty score; constructing a semantic vector, and integrating the sentence-level comprehensive loyalty to obtain a document-level semantic vector; document aggregation is carried out, and entity-level text semantic vector representation is generated; structured task integration is carried out, processing is carried out through a multi-layer sensing network, and task output is completed. According to the method and the device, the structural supervision on the output of the black box model is realized through the structured reasoning guidance and explicit loyalty feedback mechanism, the problems that the logic consistency is insufficient and the semantic credibility is difficult to quantify in the process of processing the unstructured text by the generative model are solved, and the controllability and the interpretability of the generative model are enhanced.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, specifically to a generative AI text reasoning and representation method based on a fidelity perception mechanism. Background Technology

[0002] In recent years, with the development of generative artificial intelligence (Gen AI) and large language models, they have demonstrated powerful capabilities in text generation, understanding, and reasoning. Especially when processing unstructured text data, generative AI models can automatically generate interpretive analysis content in the absence of a clear structure, providing auxiliary information for downstream tasks.

[0003] In related technologies, to enhance the reasoning ability and transparency of models, existing techniques have introduced a "Chain-of-Thought (CoT)" prompting mechanism. This mechanism guides the model to generate answers step-by-step along a logical path through a predefined series of reasoning steps. This type of mechanism has been used in some mathematical problem-solving and question-answering systems, demonstrating good performance in improving reasoning accuracy. However, most of these methods only focus on the logical coherence of the output. Since the generation behavior of large language models is still based on probabilistic language modeling, there is a risk that the actual reasoning path deviates from the explicit prompting steps; that is, "the surface logic is consistent, but the internal decision-making mechanism is opaque," leading users to mistakenly trust the model's judgment.

[0004] Other methods attempt to extract sentence embeddings or semantic representations from the output content as feature inputs for downstream analysis. For example, they use models such as BERT and RoBERTa to encode text and then directly aggregate the representation for classification or ranking. However, these methods typically ignore the position and credibility of each segment of the generated text in the logical chain, failing to distinguish which text sentences are highly faithful inference products and which may deviate from the inference chain. Ultimately, this results in an overall representation that lacks controllability and reliability, affecting the performance and interpretability of subsequent models.

[0005] Therefore, existing generative models mostly rely on implicit statistical patterns in deep neural networks for generation, lacking a controllable logical structure. This results in opaque, unreliable, and even logically deviating content from the intended goal. Generative models also suffer from insufficient logical consistency and difficulty in quantifying semantic credibility when processing unstructured text. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this application provides a generative AI text reasoning and representation method based on a fidelity-aware mechanism, which solves the problems of insufficient logical consistency and difficulty in quantifying semantic credibility in generative models when processing unstructured text.

[0007] To achieve the above objectives, this application provides the following technical solution: In a first aspect, embodiments of this application provide a generative AI text reasoning and representation method based on a fidelity-aware mechanism. This method includes: constructing a complete thought chain; guiding the construction of generative analysis content based on the complete thought chain; segmenting the generative analysis content to obtain segmented content; progressively truncating the complete thought chain to obtain multi-level truncated thought chains; collecting multiple AI analysis contents generated by the truncated thought chains; measuring the fidelity of each sentence in the generative analysis content constructed based on the complete thought chain; and constructing a fidelity scoring mechanism based on semantic similarity to measure the fidelity of each sentence in the complete thought chain. The logical consistency of content generated under the thought chain is used to calculate the sentence-level comprehensive fidelity score. A pre-trained BERT model is used to construct the semantic vectors of complete thought chain sentences, and these sentence-level comprehensive fidelity scores are integrated to obtain a document-level semantic vector with fused fidelity. A sequence modeling mechanism based on a BiGRU structure is employed, combined with a global attention module to assign different weights to different generated content, and document aggregation is performed to generate a unified entity-level text semantic vector representation. The entity-level text semantic vector representation is then integrated with the target structured data for a structured task, and processed through a multilayer perceptron (MLP) network to complete the task output. The target structured data corresponds to the research subject information.

[0008] According to a first aspect of the embodiments of this application, the aforementioned construction of a complete thought chain, the guidance of generative analysis content based on the complete thought chain, and the segmentation of the generative analysis content to obtain segmented content may specifically include the following steps: obtaining unstructured raw text information and research subject information corresponding to the raw text information, with one research subject corresponding to one entity; inputting the raw text information and research subject information together into a generative artificial intelligence model; combining the actual needs of the task scenario with domain expert knowledge, decomposing each sub-problem corresponding to the actual needs, and establishing a complete thought chain; the complete thought chain consists of... The reasoning steps consist of a set of steps and satisfy the expression: , Indicates the first The process involves several reasoning steps; guided by prompts based on a complete thought chain, it generates generative analysis content for analyzing text content. This generative analysis content is progressively reasoning-based and can represent the characteristics of the research subject and subject-related content; the generative analysis content is then segmented at the sentence level, with each document being divided into... Each sentence is segmented to obtain its content for further sentence-level semantic representation. It is a positive integer.

[0009] According to a first aspect of the embodiments of this application, the aforementioned stepwise truncation of the complete thought chain to obtain multi-level truncated thought chains, and collection of multiple AI analysis contents generated by the truncated thought chains, measuring the fidelity of each sentence in the generative analysis contents constructed based on the complete thought chain, specifically may include the following steps: sequentially truncating each reasoning step of the complete thought chain to generate k truncated thought chains of different lengths; wherein, each truncated thought chain uses the same input and target task as the complete thought chain; based on the same input and target task, using the truncated thought chains to guide GenAI to generate corresponding AI analysis contents; calculating the semantic similarity between each sentence corresponding to the generative analysis contents determined by the complete thought chain and the AI ​​analysis contents corresponding to different truncated versions of the truncated thought chains; using the BERT-based sentence-level embedding model SBERT to generate embedding vectors, and using the cosine similarity index for comparison to obtain semantic similarity information to measure the fidelity of each sentence in the generative analysis contents.

[0010] According to a first aspect of the embodiments of this application, the aforementioned k truncated thought chains include: a first truncated thought chain containing a single reasoning step. The second truncated thought chain contains two reasoning steps. , and the (i-1)th truncated thought chain containing i-1 reasoning steps.

[0011] According to a first aspect of the embodiments of this application, the aforementioned fidelity scoring mechanism based on semantic similarity measures the logical consistency of the content generated by each sentence under the complete thought chain, in order to calculate the sentence-level comprehensive fidelity score, including: calculating the complement value, summing and averaging the semantic similarity information corresponding to k truncated thought chains in sequence to obtain the sentence-level comprehensive fidelity score of each sentence in the AI ​​analysis content, so as to measure the logical consistency of the content generated by each sentence under the complete thought chain.

[0012] According to a first aspect of the embodiments of this application, the aforementioned construction of semantic vectors for complete thought chain sentences using a pre-trained BERT model, and the integration of sentence-level comprehensive fidelity scores to obtain document-level semantic vectors with fused fidelity, may specifically include the following steps: using a pre-trained BERT model for a specific text type to construct semantic vectors for each sentence to obtain document embedding vectors for the research subject; performing a dot product between the sentence-level comprehensive fidelity score and the corresponding sentence's semantic vector to obtain a sentence-level vector with fused fidelity perception mechanism, thereby integrating the sentence-level comprehensive fidelity score as an attention factor into the text vector representation to achieve dynamic adjustment of the weights of different sentences; summing and averaging multiple sentence-level vectors with fused fidelity perception mechanisms to obtain document-level semantic vectors with fused fidelity, thereby optimizing the credibility of the overall text representation.

[0013] According to a first aspect of the embodiments of this application, the aforementioned sequence modeling mechanism based on a BiGRU structure, combined with a global attention module to assign different weights to different generated content, performs document aggregation, and generates a unified entity-level text semantic vector representation, specifically including the following steps: using a sequence modeling mechanism based on a BiGRU structure, based on a preset BiGRU module, multiple document-level semantic vectors are processed through GRU units in two different directions, integrating contextual information into the document-level semantic representation to obtain document-level input information; adaptively assigning importance weights to each text through a global attention mechanism, inputting the document-level input information into a linear layer, and obtaining the input of candidate states through a non-linear activation function; normalizing the attention score using a softmax function, and obtaining an entity-level text semantic vector representation by weighted averaging of all document-level inputs.

[0014] According to a first aspect of the embodiments of this application, a structured task is performed to integrate entity-level text semantic vector representation with target structured data, and the data is processed through a multilayer perceptron (MLP) to complete the task output, including: concatenating entity-level text semantic vector representation with target structured data including statistical indicators and quantitative features as input. The stitched information is input into a multilayer perceptron (MLP) for processing, completing the final task output; the aforementioned processing steps of the MLP satisfy the expression: in, , , , , and For training parameters; For the target structured data, It is a semantic vector representation of entity-level text; and For two different intermediate values, and All are activation functions; This indicates the final task output.

[0015] Secondly, embodiments of this application provide a generative AI text reasoning and representation system based on a fidelity-aware mechanism. This generative AI text reasoning and representation system based on a fidelity-aware mechanism includes: a content segmentation module, a first measurement module, a second measurement module, a fidelity fusion module, a document aggregation module, and a task integration and output module.

[0016] Specifically, the content segmentation module constructs a complete thought chain, guides the construction of generative analysis content based on the complete thought chain, and segments the generative analysis content to obtain segmented content; the first measurement module progressively truncates the complete thought chain to obtain multi-level truncated thought chains, and collects multiple AI analysis contents generated by the truncated thought chains, measuring the fidelity of each sentence in the generative analysis content constructed based on the complete thought chain; the second measurement module constructs a fidelity scoring mechanism based on semantic similarity, measures the logical consistency of each sentence in the content generated under the complete thought chain, and calculates a sentence-level comprehensive fidelity score; the fidelity fusion module... The first module is used to construct semantic vectors for complete thought chain sentences using a pre-trained BERT model, and integrate sentence-level comprehensive fidelity scores to obtain document-level semantic vectors with fused fidelity. The second module uses a BiGRU-based sequence modeling mechanism, combined with a global attention module, to assign different weights to different generated content, perform document aggregation, and generate a unified entity-level text semantic vector representation. The third module integrates the entity-level text semantic vector representation with the target structured data for structured task integration, and processes it through a multilayer perceptron (MLP) network to complete the task output. The target structured data corresponds to the research subject information.

[0017] Thirdly, embodiments of this application provide an electronic device, which includes: a processor, a memory, and a program stored in the memory and executable on the processor. When the program is executed by the processor, it implements the generative AI text reasoning and representation method based on the fidelity perception mechanism described in the first aspect above.

[0018] Fourthly, embodiments of this application provide a computer-readable storage medium storing a program or instructions that, when executed by a processor, implement the generative AI text reasoning and representation method based on a fidelity-aware mechanism described in the first aspect.

[0019] This application provides a generative AI text reasoning and representation method based on a fidelity-aware mechanism. Compared with existing technologies, it has the following advantages: In generative AI text reasoning and representation analysis, this application constructs a complete thought chain to effectively identify whether AI-generated content truly follows the designed reasoning path and reduce the risk of illusion. This complete thought chain is then progressively truncated to obtain multi-level truncated thought chains. Based on the complete and truncated thought chains, fidelity is measured. First, a fidelity scoring mechanism is used to calculate a sentence-level comprehensive fidelity score, enhancing the logical consistency and credibility of the generated content. Then, a pre-trained BERT model is used to construct semantic vectors for sentences within the complete thought chain, integrating the sentence-level comprehensive fidelity scores to obtain document-level semantic vectors. Furthermore, different weights are assigned to different generated content, and document aggregation is performed to generate a unified entity-level text semantic vector representation. By assigning weights, high-fidelity content is enhanced in the final semantic representation, making the model more focused on reliable information and improving the task relevance and discriminative ability of the text representation. This application achieves structural supervision of the output of the black-box model through structured reasoning guidance and explicit fidelity feedback mechanism, which solves the problems of insufficient logical consistency and difficulty in quantifying semantic credibility in generative models when processing unstructured text, and enhances the controllability and interpretability of generative models. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating a generative AI text reasoning and representation method based on a fidelity-aware mechanism provided in an embodiment of this application. Figure 2 This is an exemplary detailed operation flowchart of the generative AI text reasoning and representation method based on a fidelity perception mechanism provided in the embodiments of this application; Figure 3 This is an exemplary flowchart illustrating loyalty analysis based on complete and truncated thought chains, provided in an embodiment of this application. Figure 4 This is a schematic diagram of the structure of a generative AI text reasoning and representation system based on a fidelity perception mechanism provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are described clearly and completely. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0023] This application provides a generative AI text reasoning and representation method based on a fidelity-aware mechanism, which solves the problems of insufficient logical consistency and difficulty in quantifying semantic credibility in generative models when processing unstructured text.

[0024] The technical solution in this application is to solve the above-mentioned technical problems, and the general idea is as follows: In recent years, with the development of generative artificial intelligence (Gen AI) and large language models, they have demonstrated powerful capabilities in text generation, understanding, and reasoning. Especially when processing unstructured text data, generative AI models can automatically generate interpretive analysis content in the absence of a clear structure, providing auxiliary information for downstream tasks.

[0025] In related technologies, to enhance the reasoning ability and transparency of models, existing technologies have introduced a "Chain-of-Thought (CoT)" prompting mechanism. This mechanism guides the model to generate answers step-by-step according to a logical path through a predefined series of reasoning steps. This type of mechanism has been used in some mathematical problem-solving and question-answering systems, demonstrating good performance in improving reasoning accuracy. This method guides generative language models (such as GPT) to complete the analysis task by explicitly listing multi-step reasoning processes in the prompts. For example, in complex problem-solving or logical judgment tasks, prompt structures such as "Step 1: Analyze…, Step 2: Judge…, Step 3: Draw a conclusion" can be constructed to guide the model to generate reasoning content step by step. Although this method improves the logical coherence of the generated text in form, it still has two key drawbacks: First, current mainstream generative models do not possess the ability to actually execute logical paths; their generation process is still based on language probability modeling, and the output reasoning steps may not be the model's actual internal cognitive path. Second, existing technologies lack a systematic mechanism to measure the degree to which the model's generated content follows the pre-defined reasoning path, making it impossible to determine whether the generated analysis is truly "faithful" to the logical framework in the prompts. This potential deviation may result in generated content that appears reasonable and structurally complete on the surface, but actually contains inaccurate information or flawed reasoning, thus misleading users.

[0026] Other methods attempt to extract sentence embeddings or semantic representations from the output content as feature inputs for downstream analysis. These methods typically take the original text input, such as encoders like BERT or RoBERTa, and extract sentence or document-level embedding vectors as semantic features for subsequent classification, prediction, or ranking tasks. For example, in text classification, representation vectors can be extracted and input into shallow neural networks or logistic regression models to determine the category or risk level of the text. However, these methods have significant shortcomings: First, the model does not perform structural logical analysis of the text content, and the semantic representation cannot reflect the reasoning relationships between information; second, the generated feature vectors lack interpretability, making it impossible to determine whether the model has captured the key semantics relevant to the task, especially in high-risk application scenarios where the reliability of the output is difficult to ensure.

[0027] Therefore, existing generative models mostly rely on implicit statistical patterns in deep neural networks for generation, lacking a controllable logical structure. This results in opaque, unreliable, and even logically deviating content from the intended goal. Generative models also suffer from insufficient logical consistency and difficulty in quantifying semantic credibility when processing unstructured text.

[0028] In summary, existing technologies for processing complex text using generative AI generally suffer from two types of problems: first, the lack of logical structure guidance leads to a lack of depth and transparency in the model's understanding of unstructured text; second, the lack of a mechanism to measure the reliability of generated reasoning content, making it particularly difficult to determine whether the model truly follows the specified thought chain in its reasoning. To address these issues, this invention proposes a generative AI text processing method that integrates logical guidance and semantic fidelity control mechanisms to improve the controllability, credibility, and semantic quality of generated content.

[0029] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0030] The following section first introduces a generative AI text reasoning and representation method based on a fidelity perception mechanism provided in the embodiments of this application.

[0031] This application provides a flowchart illustrating a generative AI text reasoning and representation method based on a fidelity-aware mechanism, as shown in the embodiments below. Figure 1 As shown, the generative AI text reasoning and representation method based on the fidelity perception mechanism may include the following steps S110-S160.

[0032] S110. Construct a complete thought chain, guide the construction of generative analysis content based on the complete thought chain, and segment the generative analysis content to obtain segmented content; S120. The complete thought chain is progressively truncated to obtain a multi-level truncated thought chain, and multiple AI analysis contents generated by the truncated thought chains are collected. The fidelity of each sentence in the generative analysis contents built based on the complete thought chain is measured. S130. Construct a loyalty scoring mechanism based on semantic similarity to measure the logical consistency of each sentence in the generated content under the complete thought chain, so as to calculate the sentence-level comprehensive loyalty score. S140. Construct the semantic vector of the complete thought chain sentence through the pre-trained BERT model, and integrate the sentence-level comprehensive fidelity score to obtain the document-level semantic vector with fused fidelity. S150. A sequence modeling mechanism based on BiGRU structure is adopted, and a global attention module is used to assign different weights to different generated content to perform document aggregation and generate a unified entity-level text semantic vector representation. S160. The entity-level text semantic vector representation and the target structured data are integrated into a structured task and processed through a multilayer perceptron (MLP) to complete the task output; wherein, the target structured data corresponds to the research subject information.

[0033] The above are specific implementation methods of the generative AI text reasoning and representation method based on a fidelity-aware mechanism provided in this application. Please refer to them as well. Figure 1 and Figure 2 This application provides a systematic mechanism that combines reasoning chain guidance, fidelity assessment, and semantic representation adjustment. In generative AI text reasoning and representation analysis, in order to effectively identify whether AI-generated content truly follows the designed reasoning path and reduce the risk of illusion, a complete thought chain is constructed. At the same time, the complete thought chain is gradually truncated to obtain multi-level truncated thought chains. Fidelity is measured based on the complete thought chain and the truncated thought chain.

[0034] Specifically, this application focuses on three stages in its analysis of the research subject: sentence-level comprehensive fidelity scoring, document-level semantic vectors, and entity-level text semantic vector representation. First, it calculates sentence-level comprehensive fidelity scores using a fidelity scoring mechanism to enhance the logical consistency and credibility of the generated content. Then, it constructs semantic vectors for complete thought chain sentences using a pre-trained BERT model, integrates the sentence-level comprehensive fidelity scores, and obtains document-level semantic vectors through fidelity fusion. Furthermore, it assigns different weights to different generated content, performs document aggregation, and generates a unified entity-level text semantic vector representation. By allocating weights, high-fidelity content is enhanced in the final semantic representation, making the model more focused on reliable information and improving the task relevance and discriminative ability of the text representation.

[0035] Based on this, this application achieves structured supervision of the black-box model output through structured reasoning guidance and explicit fidelity feedback mechanisms. This solves the problems of insufficient logical consistency and difficulty in quantifying semantic credibility in generative models when processing unstructured text, enhancing the controllability and interpretability of generative models. This method is applicable to various scenarios involving generative AI text analysis, such as financial assessment, legal assistance, and medical document interpretation.

[0036] This application performs structural logical analysis on text content, revealing the reasoning relationships between information. The generated feature vectors (e.g., entity-level text semantic vectors) are interpretable, capable of determining whether key task-related semantics have been captured, ensuring output reliability, especially in high-risk application scenarios. On one hand, it possesses the ability to realistically execute logical paths; its generation process is not solely based on language probabilistic modeling, and the output reasoning steps correspond to the model's actual internal cognitive path. On the other hand, it can also measure the degree to which the model-generated content adheres to the preset reasoning path, determining whether the generation analysis truly "faithfully" follows the logical framework in the prompt. This application can be widely applied to scenarios such as semantic understanding, automatic reasoning, feature extraction, and task decision-making for unstructured text information, possessing broad deployment value and good scalability and applicability.

[0037] In some embodiments, the aforementioned construction of a complete thought chain, the construction of generative analysis content based on the complete thought chain, and the segmentation of the generative analysis content to obtain segmented content, that is, the aforementioned S110 may specifically include the following steps: S210. Obtain unstructured raw text information and research subject information corresponding to the raw text information, with one research subject corresponding to one entity; S220. Input the original text information and the research subject information into the generative artificial intelligence model; S230. Combining the actual needs of the task scenario with domain expert knowledge, decompose the sub-problems corresponding to the actual needs and establish a complete thought chain; the complete thought chain consists of... The reasoning steps consist of a set of steps and satisfy the expression: , Indicates the first One reasoning step; S240. Based on prompts and guidance of the complete thought chain, generative analysis content is generated for analyzing text content. The generative analysis content is progressive reasoning content and can represent the characteristics of the research subject and subject-related content. S250. Perform sentence-level segmentation on the generative analysis content, and each document is segmented into... Each sentence is segmented to obtain its content for further sentence-level semantic representation. It is a positive integer.

[0038] In the embodiments of this application, it is understood that this application obtains unstructured raw text information, such as industry announcements, business records, or public documents, along with research subject information, and inputs them together into the generative artificial intelligence model GenAI. Under the guidance of a complete, structured thought chain designed by experts, it generates step-by-step reasoning content for analyzing the text content. , This represents a thought process chain; it reflects the characteristics of the research subject and related content, and segments the content into sentences, denoted as... Each document is therefore divided into These sentences are used for further sentence-level semantic representation. This represents the first sentence obtained from the segmentation. This represents the second sentence obtained from the segmentation. This represents the nth sentence obtained from the segmentation. This represents the set of n sentences obtained from the segmentation. The complete thought chain design includes multiple logical steps, covering tasks such as semantic extraction, comparison and judgment, and conclusion summarization.

[0039] It should be noted that this application designs a multi-step thought chain prompting structure jointly guided by domain expert knowledge and task requirements. This structure explicitly breaks down complex problems into progressive logical reasoning tasks, guiding generative AI (such as large language models) to gradually perform information extraction, content comparison, and comprehensive judgment. This structured prompting differs from general prompts or open-ended dialogue generation, significantly improving the transparency of the model's reasoning path and the completeness of the analyzed content.

[0040] In some embodiments, the aforementioned complete thought chain is progressively truncated to obtain multi-level truncated thought chains, and multiple AI analysis contents generated by the truncated thought chains are collected. The fidelity of each sentence in the generative analysis content constructed based on the complete thought chain is measured. That is, the aforementioned S120 may specifically include the following steps: S310. Each reasoning step of the complete thought chain is truncated sequentially to generate k truncated thought chains of different lengths; wherein each truncated thought chain uses the same input and target task as the complete thought chain. S320. Based on the same input and target task, GenAI is guided to generate corresponding AI analysis content using the truncated thought chain. S330. Calculate the semantic similarity between each sentence corresponding to the generative analysis content determined by the complete thought chain and the AI ​​analysis content corresponding to the truncated thought chain of different truncated versions. S340. The sentence-level embedding model SBERT based on BERT is used to generate embedding vectors, and cosine similarity index is used to compare them to obtain semantic similarity information to measure the fidelity of each sentence in the generative analysis content.

[0041] For example, the aforementioned k truncated thought chains include: a first truncated thought chain containing a single reasoning step. The second truncated thought chain contains two reasoning steps. , and the (i-1)th truncated thought chain containing i-1 reasoning steps.

[0042] In the embodiments of this application, it is understood that, please refer to Figure 3 This application constructs a complete thought chain, consisting of... It consists of several reasoning steps, and through truncation, a total of k truncated thought chains are obtained. , Let represent the k-th truncated thought chain. This application then collects AI analysis content generated by different truncated thought chains: based on the same original input and target task, these truncated thought chains guide GenAI to generate corresponding analysis content. Each truncated thought chain generates corresponding analysis content based on m sets of unstructured original text information. ;in, This represents the analysis content generated from m sets of unstructured raw text information for the k-th truncated thought chain.

[0043] Furthermore, this application measures the fidelity of each sentence based on the complete thought chain: calculating the fidelity of each sentence in the complete CoT-generated analysis content. To analyze the semantic similarity between documents with different truncated versions, an embedding vector was generated using a BERT model (denoted as SBERT) that can convert text into sentence vectors, and the cosine similarity index was used for comparison.

[0044] In the formula, This represents the first step in the generation of a complete thought chain. The first AI analysis content The sentence and the first The first generation of the truncated thought chain guided by the generation AI analyzes the semantic similarity between content.

[0045] It should be noted that this application provides a multi-version comparison mechanism based on truncated thought chains. To measure whether AI-generated content truly follows the preset reasoning chain logic, this application constructs multiple "truncated versions" of thought chains based on a complete thought chain design, with each version containing only a portion of the steps of the original reasoning chain. By generating multiple versions of analysis content for the same input text under thought chain prompts of different lengths, a comparison system based on "reasoning path integrity" is constructed, providing a basic sample for subsequent fidelity assessment.

[0046] In some embodiments, the aforementioned fidelity scoring mechanism based on semantic similarity measures the logical consistency of each sentence in the content generated under the complete thought chain, in order to calculate the sentence-level comprehensive fidelity score. Specifically, S130 may include: calculating the complement value, summing and averaging the semantic similarity information corresponding to k truncated thought chains in turn to obtain the sentence-level comprehensive fidelity score of each sentence in the AI ​​analysis content, in order to measure the logical consistency of each sentence in the content generated under the complete thought chain.

[0047] In the embodiments of this application, it is understood that low similarity represents higher uniqueness and reasoning depth. This application calculates fidelity by taking the complement values ​​of the similarity scores on k truncated thought chains, summing them, and taking the average. Correspondingly, the first thought chain guides the generation of the complete thought chain. The first AI analysis content Overall fidelity score of each sentence as follows: In the formula, In the formula, This represents the result obtained by iterating through all the sentences corresponding to the m pieces of analysis content, that is, the total number of sentences included in the m pieces of analysis content. Each sentence has a sentence-level comprehensive fidelity score, which will be integrated into the text vector representation as an attention factor to achieve dynamic adjustment of the weights of different sentences.

[0048] It should be noted that this application constructs a sentence-level fidelity scoring mechanism to measure the logical consistency of each sentence in the generated content under the complete thought chain. In the generated complete analysis text, this application segments the content by sentence and compares the changes in similarity between each sentence under the complete thought chain and each truncated thought chain. Sentences that are relatively difficult to generate from simplified prompts are considered to have higher "logical dependence" and "reasoning fidelity"; this mechanism effectively filters out the most representative and credible linguistic information in the reasoning path.

[0049] In some embodiments, the aforementioned construction of the semantic vector of a complete thought chain sentence using a pre-trained BERT model, and integration of sentence-level comprehensive fidelity scores, yields a document-level semantic vector with fused fidelity. Specifically, S140 may include the following steps: S410. Using a pre-trained BERT model for a specific text type, construct semantic vectors for each sentence to obtain the document embedding vector of the research subject. S420. Based on the sentence-level comprehensive loyalty score, a dot product is performed with the semantic vector of the corresponding sentence to obtain a sentence-level vector that integrates the loyalty perception mechanism. This allows the sentence-level comprehensive loyalty score to be integrated as an attention factor into the text vector representation, enabling dynamic adjustment of the weights of different sentences. S430. Sum and average the sentence-level vectors of multiple fusion fidelity perception mechanisms to obtain the document-level semantic vector of fusion fidelity, thereby optimizing the credibility of the overall text representation.

[0050] In the embodiments of this application, it can be understood that this application uses a pre-trained language model (denoted as XBERT) for a specific text type to construct semantic vectors for each sentence.

[0051] make This indicates that a research subject produces [something] sequentially over a period of time (e.g., one year). The document embedding vectors of the reports, where each report is composed of... The system consists of several sentences, each semantically represented using a language model. Then, the overall loyalty score is combined with the corresponding sentence's semantic vector and multiplied by a dot product to obtain a sentence-level vector that incorporates the loyalty-aware mechanism.

[0052] In the formula, This represents a sentence-level representation that incorporates fidelity scores. A higher fidelity score indicates that the sentence guided by the complete thought chain has higher fidelity. The greater the impact on model predictions.

[0053] This application provides a sentence-level text vector that incorporates a complete thought chain-guided AI-generated analysis content fidelity. Summation and averaging are performed to obtain the document-level semantic vector with fused fidelity. Sentence-level vectors are aggregated to generate document-level semantic representations, thereby optimizing the credibility of the overall text representation. Document-level semantic vectors Satisfying the expression: It should be noted that this application constructs a fidelity-aware text vector enhancement mechanism. This mechanism uses the aforementioned sentence-level comprehensive fidelity score as a weighting factor, integrating it into the semantic vector of each sentence. Unlike traditional models that directly aggregate all text content, this mechanism emphasizes the contribution of highly fidelity sentences while downplaying content with unclear or potentially deviating reasoning paths, thereby generating a more logically interpretable and reliable overall text representation. This mechanism can be widely embedded into representation learning modules for various natural language processing tasks.

[0054] In some embodiments, the aforementioned sequence modeling mechanism based on the BiGRU structure, combined with a global attention module, assigns different weights to different generated content, performs document aggregation, and generates a unified entity-level text semantic vector representation. That is, the aforementioned S150 may specifically include the following steps: S510. A sequence modeling mechanism based on BiGRU structure is adopted. Based on the preset BiGRU module, multiple document-level semantic vectors are processed through GRU units in two different directions, and the context information is integrated into the document-level semantic representation to obtain document-level input information. S520: Adaptively assign importance weights to each text using a global attention mechanism, input document-level input information into a linear layer, and obtain the input of candidate states through a non-linear activation function; S530. The attention score is normalized using the softmax function. The entity-level text semantic vector representation is obtained by weighted averaging of all document-level inputs.

[0055] In the embodiments of this application, it is understood that, firstly, for situations where there are multiple related texts (such as multiple rounds of analysis or multiple document inputs) and the texts exhibit temporal sequence characteristics and heterogeneous importance, a sequence modeling mechanism based on a BiGRU structure is adopted. This is combined with a global attention module to assign different weights to different generated content, thereby generating a unified entity-level text semantic vector representation. The BiGRU module integrates contextual information into the document-level semantic representation through two GRU units in different directions, thus obtaining the document-level input information. The process can be represented as follows: The global attention mechanism adaptively assigns importance weights to each text element. Specifically, it considers document-level input information. The input is first fed into a linear layer, where it passes through a non-linear activation function to obtain the candidate state input. Subsequently, the attention score is normalized using the softmax function. Finally, a weighted average of all document-level inputs is used to obtain the entity-level text semantic vector representation. .

[0056] It should be noted that this application constructs a structured-unstructured fusion representation modeling and prediction mechanism. In downstream tasks, this application jointly models the enhanced text representation with structured data (such as tabular data, time series indicators, etc.). Sequence neural networks with attention mechanisms (such as BiGRU+Attention) can be used to further achieve dynamic weighted processing of different time points and different text types, enabling the model to more accurately grasp the text signals at key time points, and finally output task results (such as classification probabilities or decision judgments) through a fully connected prediction module. This structure achieves reliable control over AI-generated text, unified expression of diverse information sources, and the ability to model the temporal evolution of text.

[0057] In some embodiments, the aforementioned integration of entity-level text semantic vector representation with target structured data into a structured task, and processing through a multilayer perceptron (MLP) network to complete the task output, specifically includes the following steps in S160: S610. Input the entity-level text semantic vector representation by concatenating it with the target structured data, which includes statistical indicators and quantitative features. S620. Input the stitched information into the multilayer perceptron (MLP) for processing and complete the final task output. The processing steps of the aforementioned multilayer perceptron (MLP) network satisfy the expression: in, , , , , and For training parameters; For the target structured data, It is a semantic vector representation of entity-level text; and For two different intermediate values, and All are activation functions; This indicates the final task output.

[0058] In some embodiments, this application provides a generative AI text reasoning and representation system 700 based on a fidelity-aware mechanism, such as... Figure 4 As shown, the generative AI text reasoning and representation system 700 based on a fidelity-aware mechanism may include the following modules: The content segmentation module 710 is used to construct a complete thought chain, guide the construction of generative analysis content based on the complete thought chain, and segment the generative analysis content to obtain segmented content. The first measurement module 720 is used to progressively truncate the complete thought chain to obtain a multi-level truncated thought chain, and collect multiple AI analysis contents generated by the truncated thought chain. The fidelity of each sentence in the generative analysis content constructed based on the complete thought chain is measured. The second measurement module 730 is used to build a loyalty scoring mechanism based on semantic similarity, measure the logical consistency of the content generated by each sentence in the complete thought chain, and calculate the sentence-level comprehensive loyalty score. The fidelity fusion module 740 is used to construct the semantic vector of the complete thought chain sentence through the pre-trained BERT model, and integrate the sentence-level comprehensive fidelity score to obtain the document-level semantic vector with fused fidelity. The document aggregation module 750 is used to perform document aggregation by using a sequence modeling mechanism based on the BiGRU structure, combined with a global attention module to assign different weights to different generated content, and generate a unified entity-level text semantic vector representation. The task integration output module 760 is used to integrate entity-level text semantic vector representations with target structured data in a structured task, and process the data through a multilayer perceptron (MLP) to complete the task output; wherein the target structured data corresponds to the research subject information.

[0059] According to embodiments of this application, any and multiple modules among the content segmentation module 710, the first measurement module 720, the second measurement module 730, the fidelity fusion module 740, the document aggregation module 750, and the task integration output module 760 can be merged into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module.

[0060] Figure 4 Each module in the system shown has the function of implementing each step in the aforementioned generative AI text reasoning and representation method based on the fidelity perception mechanism, and can achieve its corresponding technical effect. For the sake of brevity, it will not be elaborated here.

[0061] In some embodiments, this application provides an electronic device, the structural schematic of which is shown below. Figure 5 As shown.

[0062] The electronic device may include a processor 810 and a memory 820 storing computer program instructions.

[0063] Specifically, the processor 810 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0064] Memory 820 may include mass storage for data or instructions. For example, and not limitingly, memory 820 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 820 may include removable or non-removable (or fixed) media. Where appropriate, memory 820 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 820 is non-volatile solid-state memory.

[0065] Memory 820 may include read-only memory (ROM), random access memory (RAM), disk storage media device, optical storage media device, flash memory device, electrical, optical, or other physical / tangible memory storage device. Therefore, typically, memory 820 includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it can perform the operations described in any of the generative AI text reasoning and representation methods based on fidelity-aware mechanisms in the above embodiments.

[0066] The processor 810 reads and executes computer program instructions stored in the memory 820 to implement any of the generative AI text reasoning and representation methods based on the fidelity perception mechanism in the above embodiments.

[0067] In one example, the electronic device may also include a communication interface 830 and a bus 800. For example, Figure 5 As shown, the processor 810, memory 820, and communication interface 830 are connected via bus 800 and communicate with each other.

[0068] The communication interface 830 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application. Bus 800 includes hardware, software, or both, that couples components of an online data traffic metering device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 800 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.

[0069] Furthermore, in conjunction with the generative AI text reasoning and representation method based on a fidelity-aware mechanism in the above embodiments, this application embodiment can provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the generative AI text reasoning and representation methods based on a fidelity-aware mechanism in the above embodiments.

[0070] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0071] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0072] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0073] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0074] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generative AI text reasoning and representation based on loyalty awareness mechanism, characterized in that, The method comprises the following steps: constructing a complete thought chain, guiding the construction of generative analysis content based on the complete thought chain, and segmenting the generative analysis content to obtain segmented content; gradually truncating the complete thought chain to obtain multiple levels of truncated thought chains, collecting multiple AI analysis contents generated by the truncated thought chains, and measuring the fidelity of each sentence in the generative analysis content based on the complete thought chain; building a fidelity scoring mechanism based on semantic similarity to measure the logical consistency of each sentence in the generated content under the complete thought chain to calculate a sentence-level comprehensive fidelity score; building a semantic vector of the complete thought chain sentence by pre-training a BERT model, and integrating the sentence-level comprehensive fidelity score to obtain a document-level semantic vector with fused fidelity; adopting a sequence modeling mechanism based on a BiGRU structure, combining a global attention module to assign different weights to different generated content, performing document aggregation, and generating a unified entity-level text semantic vector representation; integrating the entity-level text semantic vector representation with target structured data for structured task integration, and processing through a multi-layer perception network MLP to complete task output; wherein the target structured data corresponds to research subject information.

2. The loyalty-awareness mechanism based generative AI text reasoning and representation method of claim 1, wherein, The method comprises the following steps: obtaining unstructured original text information and research subject information corresponding to the original text information, one research subject corresponding to one entity; inputting the original text information and the research subject information into a generative artificial intelligence model; Combining the actual needs of the task scenario with domain expert knowledge, we break down the sub-problems corresponding to the actual needs and establish a complete thought chain; the complete thought chain consists of... The reasoning steps consist of a set of steps and satisfy the expression: , Indicates the first One reasoning step; generating generative analysis content for analyzing text content based on the prompt guidance of the complete thought chain, the generative analysis content being step-by-step reasoning content and being able to represent research subject characteristics and subject-related content; The generative analysis content is sentence-level segmented, each document is segmented into a plurality of sentences, obtaining segmented content for further sentence-level semantic representation, is a positive integer.

3. The loyalty-awareness mechanism based generative AI text reasoning and representation method of claim 2, wherein, The method comprises the following steps: sequentially truncating each reasoning step of the complete thought chain to generate k truncated thought chains of different lengths; wherein each truncated thought chain uses the same input and target task as the complete thought chain; generating corresponding AI analysis content using the truncated thought chain based on the same input and target task; calculating the semantic similarity between each sentence of the generative analysis content determined by the complete thought chain and the AI analysis content corresponding to the truncated thought chain of different truncated versions; generating embedding vectors using a BERT-based sentence-level embedding model SBERT, and comparing using a cosine similarity index to obtain semantic similarity information to measure the fidelity of each sentence in the generative analysis content.

4. The loyalty-awareness mechanism based generative AI text reasoning and representation method of claim 3, wherein, The method comprises the following steps: The semantic similarity information corresponding to the k truncated thought chains is sequentially calculated for complement, summation and average to obtain a sentence-level comprehensive fidelity score of each sentence in the AI analysis content, so as to measure the logical consistency of each sentence in the generated content under the complete thought chain; The k truncated thought chains comprise a first truncated thought chain comprising a single inference step a second truncated thought chain comprising two inference steps and an i-1 truncated thought chain comprising i-1 inference steps.

5. The loyalty-awareness mechanism based generative AI text reasoning and representation method of claim 1, wherein, The semantic vector of the sentence of the complete thought chain is constructed by the pre-trained BERT model, and the document-level semantic vector of the fusion fidelity is obtained by integrating the sentence-level comprehensive fidelity score, including: Using the pre-trained BERT model for a specific text type, the semantic vector of each sentence is constructed to obtain the document embedding vector of the research subject; Based on the sentence-level comprehensive fidelity score, the dot product of the semantic vector of the corresponding sentence is performed to obtain the sentence-level vector of the fusion fidelity awareness mechanism, so as to integrate the sentence-level comprehensive fidelity score into the text vector representation as an attention factor, and realize the dynamic adjustment of the weight of different sentences; The document-level semantic vector of the fusion fidelity is obtained by summing and averaging the sentence-level vectors of multiple fusion fidelity awareness mechanisms, so as to realize the credibility optimization of the overall representation of the text.

6. The loyalty-awareness mechanism based generative AI text reasoning and representation method of claim 1, wherein, The sequence modeling mechanism based on the BiGRU structure is adopted, and the global attention module is combined to assign different weights to different generated content, aggregate the document, and generate a unified entity-level text semantic vector representation, including: The sequence modeling mechanism based on the BiGRU structure is adopted, and the BiGRU module is based on the preset BiGRU module, which processes multiple document-level semantic vectors through two different direction GRU units, integrates the context information into the document-level semantic representation, and obtains the document-level input information; Through the global attention mechanism, each text is adaptively assigned an importance weight, the document-level input information is input into a linear layer, and the input of the candidate state is obtained through a nonlinear activation function; The attention score is normalized by using the softmax function, and the entity-level text semantic vector representation is obtained by weighted averaging all document-level inputs.

7. The loyalty-awareness mechanism based generative AI text reasoning and representation method of claim 1, wherein, The entity-level text semantic vector representation is integrated with the target structured data including statistical indicators and quantitative features, and is processed by a multi-layer perception network MLP to complete the task output, including: The entity-level text semantic vector representation is integrated with the target structured data including statistical indicators and quantitative features, and is processed by a multi-layer perception network MLP to complete the task output, including: The processing steps of the multi-layer perception network MLP satisfy the expression: including: wherein, , , , , and are training parameters; is target structured data, is an entity-level textual semantic vector representation; and are two different intermediate values, and are activation functions; represents the final task output.

8. A generative AI text reasoning and representation system based on loyalty awareness mechanism, characterized in that, A content segmentation module is configured to construct a complete thought chain, guide the construction of a generative analysis content based on the complete thought chain, and segment the generative analysis content to obtain segmented content; A first measurement module is configured to truncate the complete thought chain step by step to obtain multiple levels of truncated thought chains, collect multiple AI analysis contents generated based on the truncated thought chains, and measure the fidelity of each sentence in the generative analysis content constructed based on the complete thought chain. ​ The second metric module is configured to construct a fidelity score mechanism based on semantic similarity, measure logical consistency of each sentence in the complete thought chain to generate content, and calculate a sentence-level comprehensive fidelity score; The fidelity fusion module is configured to construct a semantic vector of the complete thought chain sentence by using a pre-trained BERT model, and integrate the sentence-level comprehensive fidelity score to obtain a document-level semantic vector of the fused fidelity; The document aggregation module is configured to use a sequence modeling mechanism based on a BiGRU structure, combine a global attention module to assign different weights to different generated content, perform document aggregation, and generate a unified entity-level text semantic vector representation. The task integration output module is configured to perform structured task integration on the entity-level text semantic vector representation and target structured data corresponding to the subject information, process the target structured data through a multi-layer perception network (MLP), and complete task output.

9. An electronic device, comprising: The processor, the memory, and the program stored on the memory and executable on the processor, wherein the program is executed by the processor to implement the generative AI text reasoning and representation method based on the fidelity perception mechanism according to any one of claims 1 to 7. The computer-readable storage medium stores programs or instructions, which are executed by the processor to implement the generative AI text reasoning and representation method based on the fidelity perception mechanism according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, ​

Images