Large model analysis process generation method and device, equipment, medium and program product

Abstract

The embodiment of the invention discloses a large model analysis process generation method and device, equipment, a medium and a program product. Comprising the following steps: acquiring original process logic description information and a target user cognitive portrait, and performing structured analysis on the original process logic description information to determine structured skeleton information; according to the cognitive portrait of the target user, performing multi-level cognitive suitability evaluation on the structured skeleton information, and determining a multi-level cognitive evaluation result; wherein the hierarchy of the multi-hierarchy cognitive assessment is determined according to the hierarchy of the teaching plan structure; performing different levels of optimization on the original process logic description information according to the multi-level cognitive evaluation result and a pre-constructed optimization prompt information set, and determining a target process analysis file; wherein the optimization prompt information set comprises optimization prompt information corresponding to each level. The readability of the large model analysis process given externally is improved while the accuracy of the large model analysis process displayed externally is ensured.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method, apparatus, device, medium, and program product for generating large model analysis processes. Background Technology

[0002] In recent years, large pre-trained language models (LLMs) have rapidly developed in fields such as natural language understanding and generation, reasoning assistance, and educational tutoring, becoming a core technology for human-computer interaction and knowledge services. As model applications expand from single-turn question answering to complex tasks (such as instructional design, decision support, and legal interpretation), users' demands for the interpretability, verifiability, and teachability of model outputs are increasing. LLMs typically perform complex reasoning through coherent chains of thought, providing a technological foundation for improving model performance and bringing real-world value in translating internal reasoning into user-understandable expressions. The ability to translate model reasoning into user cognitive frameworks is becoming an important direction for promoting the practical application and widespread acceptance of artificial intelligence.

[0003] However, because the internal reasoning process of a model is usually presented in technical and fine-grained computational steps, often including a large number of symbols, assumptions, and intermediate states, it is difficult to meet the understanding needs of users with different knowledge backgrounds regarding "how the result is obtained" and "how the input is mapped to the output." At the same time, because user groups are often highly heterogeneous (e.g., educational background, logical reasoning ability, attention span, and preferred expression forms), a single technical presentation cannot accommodate these differences, leading to prioritizing the explanation of the result or imposing a cognitive burden on users. Furthermore, explaining the reasoning process solely through simple post-summaries or directly exposing the reasoning chain can easily lead to information overload, misleading information, or model illusion, failing to meet the understanding needs of users with different knowledge backgrounds regarding the internal reasoning analysis process of LLMs. Summary of the Invention

[0004] This invention provides a method, apparatus, device, medium, and program product for generating large model analysis processes. It generates stable, controllable, and interpretable expressions of the reasoning information of the internal analysis of large models in a way that is comprehensible to users. While ensuring the accuracy of the large model analysis process presented to external parties, it also improves the readability of the large model analysis process presented to external parties.

[0005] In a first aspect, embodiments of the present invention provide a method for generating a large model analysis process, including:

[0006] Obtain the original process logic description information and the target user cognitive profile, and perform structured parsing on the original process logic description information to determine the structured skeleton information;

[0007] Based on the cognitive profile of the target user, a multi-level cognitive adaptability assessment is conducted on the structured skeleton information to determine the multi-level cognitive assessment results; the levels of the multi-level cognitive assessment are determined according to the hierarchical structure of the lesson plan.

[0008] Based on the results of multi-level cognitive assessment and the pre-constructed set of optimization prompts, the original process logic description information is optimized at different levels to determine the target process analysis document; the set of optimization prompts includes optimization prompts corresponding to each level.

[0009] Secondly, embodiments of the present invention provide a large model analysis process generation apparatus, comprising:

[0010] The information structuring module is used to acquire the original process logic description information and the target user cognitive profile, and to perform structured parsing on the original process logic description information to determine the structured skeleton information;

[0011] The multi-level assessment module is used to conduct multi-level cognitive adaptability assessments on structured skeleton information based on the target user's cognitive profile, and to determine the multi-level cognitive assessment results; the levels of the multi-level cognitive assessment are determined according to the teaching plan structure levels.

[0012] The analysis file determination module is used to optimize the original process logic description information at different levels based on the results of multi-level cognitive assessment and the pre-constructed set of optimization prompts, and determine the target process analysis file; wherein, the set of optimization prompts includes optimization prompts corresponding to each level.

[0013] Thirdly, embodiments of the present invention provide a large model analysis process generation device, comprising:

[0014] At least one processor; and a memory communicatively connected to the at least one processor;

[0015] The memory stores a computer program that can be executed by at least one processor, which enables the at least one processor to implement the large model analysis process generation method of any embodiment of the present invention.

[0016] Fourthly, embodiments of the present invention also provide a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform the large model analysis process generation method of any embodiment of the present invention.

[0017] Fifthly, embodiments of the present invention also provide a computer program product, including a computer program, which, when executed by a processor, is used to perform the large model analysis process generation method of any embodiment of the present invention.

[0018] This invention provides a method, apparatus, device, medium, and program product for generating a large-scale model analysis process. It acquires original process logic description information and a target user cognitive profile, and performs structured parsing on the original process logic description information to determine structured skeleton information. Based on the target user cognitive profile, it conducts a multi-level cognitive adaptability assessment on the structured skeleton information to determine the multi-level cognitive assessment results. The levels of the multi-level cognitive assessment are determined based on the teaching plan structure levels. Based on the multi-level cognitive assessment results and a pre-constructed set of optimization prompts, the original process logic description information is optimized at different levels to determine the target process analysis file. The set of optimization prompts includes optimization prompts corresponding to each level. By adopting the above technical solution, the internal logical reasoning process of a large pre-trained language model is transformed into an analysis process that is understandable to the user, analogous to a "teaching process." After structuring the original process logic description information of the acquired large-scale pre-trained language model, based on the acquired target user cognitive profile containing user cognitive characteristics and the teaching plan structure hierarchy of the teaching plan used in the "teaching process", the original process logic description information is subjected to multi-level cognitive evaluation and optimization in accordance with the teaching plan structure hierarchy. Finally, a target process analysis document is generated that conforms to user cognitive characteristics, exists in a "teaching plan" format, and can better enable users to understand the internal logical reasoning process of the large-scale pre-trained language model. While ensuring the accuracy of the large model analysis process presented to the outside world, the readability of the large model analysis process presented to the outside world is improved.

[0019] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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 A flowchart of a method for generating a large model analysis process provided in Embodiment 1 of the present invention;

[0022] Figure 2 This is a flowchart of a method for generating a large model analysis process provided in Embodiment 2 of the present invention;

[0023] Figure 3This is an example diagram of a process structure for multi-level cognitive adaptability assessment of structured skeleton information through a lesson plan assessment module library, as provided in Embodiment 2 of the present invention.

[0024] Figure 4 This is a schematic diagram of a large model analysis process generation device provided in Embodiment 3 of the present invention;

[0025] Figure 5 This is a schematic diagram of a large model analysis process generation device provided in Embodiment 4 of the present invention. Detailed Implementation

[0026] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0027] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0028] Example 1

[0029] Figure 1 This is a flowchart illustrating a method for generating a large model analysis process according to Embodiment 1 of the present invention. This embodiment of the invention is applicable to generating process analysis documents, similar to lesson plans, that are user-understanding and provided to users to illustrate the internal logical reasoning process of a large pre-trained language model, thereby achieving a user-interpretable expression of the model's internal reasoning. This method can be executed by a large model analysis process generation device, which can be implemented by software and / or hardware and can be configured within a large model analysis process generation device. Optionally, the large model analysis process generation device can be a laptop, desktop computer, or smart tablet, etc., and this embodiment of the invention does not impose any limitations on this.

[0030] like Figure 1 As shown in the figure, the method for generating a large model analysis process provided by the embodiment of the present invention specifically includes the following steps:

[0031] S101. Obtain the original process logic description information and the target user cognitive profile, and perform structured parsing on the original process logic description information to determine the structured skeleton information.

[0032] In this embodiment, the original process logic description information can be specifically understood as the logical description information of the problem-solving process within a large pre-trained language model, or as the original inference chain used to describe the intermediate inference steps of the model. It is understood that the original process logic description information in this embodiment can be a description of the solution process logic generated by the model, or an unsystematic explanation or elaboration of a certain knowledge point, etc. It is a multimodal description in form, and its quality level is usually not high, with considerable room for optimization. This embodiment does not limit the actual content included in the original process logic description information.

[0033] In this embodiment, the target user cognitive profile can be specifically understood as a structured descriptive information used to describe the cognitive level information of the target user, wherein the target user is the user who wants to obtain the interpretable internal reasoning analysis process of a large pre-trained language model. Optionally, the target user cognitive profile may include cognitive-related features such as the target user's cognitive preferences and learning habits, which can be used to provide constraints and optimization objectives related to the user's cognitive ability for subsequent cognitive evaluation of the original process logic description information.

[0034] In this embodiment, the structured skeleton information can be specifically understood as the information obtained by multi-level structured decomposition and reorganization of the original process logic description information according to the hierarchical structure of the lesson plan, transforming the unstructured original process logic description information into a skeleton structure with a clear hierarchical relationship. It can be understood that the lesson plan structure hierarchy proposed here can be understood as the hierarchical structure commonly used in teaching, such as including three levels: whole, module, and paragraph. The whole can be specifically understood as the complete lesson plan as a whole; a module can specifically refer to different content modules in the lesson plan, such as introduction [scenario creation, review of prior knowledge, raising questions or objectives], new lesson explanation [knowledge presentation, demonstration and practice, interactive questioning, key point breakthrough, difficulty analysis], classroom exercises [basic exercises, extended exercises, immediate feedback], group activities or discussions [task allocation, collaborative inquiry, results presentation, peer review and exchange], summary and generalization [knowledge review, method summary, emotional sublimation], and homework assignment [homework type, homework requirements, next lesson preview]; a paragraph can be specifically understood as each paragraph in the lesson plan, where paragraphs may belong to different modules.

[0035] Specifically, when a user needs to obtain internal logical analysis information from a large pre-trained language model, the original process logic description information generated during the analysis process will first be obtained from the large pre-trained language model. Simultaneously, to ensure that the final displayed information better matches the user's cognitive understanding, a target user cognitive profile will be obtained. Furthermore, to better evaluate and optimize the original process logic description information, the unstructured original process logic description information needs to be structured and parsed, transforming it into a structured skeleton information with a clear hierarchical relationship, consistent with the teaching plan's structure hierarchy.

[0036] Optionally, to more clearly define the subsequent cognitive evaluation method of the target user cognitive profile on the structured skeleton information, the following will explain how to obtain the target user cognitive profile.

[0037] In some examples, a simple identity description, such as "primary school student, middle school student, university student, professional", can be used to target users. In this case, the reasoning ability of LLM can be invoked to directly generate a cognitive ability and learning preference profile for the target based on the above simple identity description, and use it as the cognitive profile of the target user.

[0038] In some examples, the target user cognitive profile for a target user can come from a complete user profiling system. The user profiling system can generate a more refined cognitive profile by deeply analyzing the historical learning data and profile information related to the user it acquires.For example, the target user's education level and domain familiarity can be converted into quantitative indicators using a scale method, with a value range of [0,1]. The model can then be trained based on the user's historical learning data to calculate values ​​for dimensions such as abstract thinking, logical reasoning, and processing speed, also with a value range of [0,1], forming a complete cognitive profile: Cognitive_profile={education_level,domain_familiarity,abstract_thinking,logical_reasoning,processing_speed,...}. A similar approach can be used to generate the target user's learning preferences. This learning preference information can include the target user's preference for text, images, cases, exercises, etc. (content_format_preference), and the appropriate information density and learning speed (lear) for the target user. Learning preferences are formed by considering factors such as learning pace, attention span, interaction style, and feedback sensitivity: Learning_preferences={content_format_preference, learning_pace, attention_span, interaction_style, feedback_sensitivity,...}. A similar method can be used to generate a target user's tolerance for content of varying complexity. This tolerance can include intrinsic cognitive load, extraneous cognitive load, related cognitive load, and fatigue accumulation pattern during continuous learning, thereby determining the target user's safe cognitive load range: Cognitive_load_limits={intrinsic_threshold, extraneous_threshold, germane_threshold, fatigue_pattern,...}.By combining the above information, the target user's cognitive profile can be determined, denoted as User_cognitive_portrait={Cognitive_profile,Learning_preferences,Cognitive_load_limits}.

[0039] It is understood that the information in the above-mentioned target user cognitive profile can be adjusted according to the actual situation, and the evaluation methods for different preference levels and load capacity can be adjusted according to the actual situation. This embodiment of the invention does not impose any restrictions on this.

[0040] S102. Based on the target user's cognitive profile, conduct a multi-level cognitive adaptability assessment of the structured skeleton information and determine the multi-level cognitive assessment results.

[0041] The levels of multi-level cognitive assessment are determined based on the hierarchical structure of the lesson plan.

[0042] In this embodiment, the multi-level cognitive assessment result can be specifically understood as the result of assessing whether the information contained in the structured skeleton information conforms to the user's cognitive ability based on the target user's cognitive profile, under multiple different teaching plan structure levels.

[0043] Specifically, since the structured skeleton information is constructed based on the hierarchical structure of the lesson plan, it can be analyzed and evaluated at different levels of the lesson plan structure. For each level of the lesson plan structure, the structured skeleton information can extract features that affect user cognition at that level, such as logical organization, information density, mnemonic design, text description, and case coverage. These features are then matched and compared with the target user's cognitive ability regarding the aforementioned features in the target user's cognitive profile. Based on the matching and comparison results, the cognitive evaluation result at that level is determined. Furthermore, the cognitive evaluation results obtained after evaluating all levels of the lesson plan structure can be combined to determine a multi-level cognitive evaluation result.

[0044] S103. Based on the results of the multi-level cognitive assessment and the pre-constructed set of optimization prompts, optimize the original process logic description information at different levels to determine the target process analysis document.

[0045] The set of optimization prompts includes optimization prompts corresponding to each level.

[0046] In this embodiment, the optimized prompt information can be specifically understood as a prompt template used to optimize the original process logic description information at different lesson plan structure levels. It can be understood that the optimized prompt information should contain one or more placeholders (prompt variables) for filling in the corresponding level's cognitive assessment results to form a complete optimized prompt information, which, together with the original process logic description information, serves as input for LLM, completing the optimization of the original process logic description information at that lesson plan level. The optimized prompt information set can be specifically understood as a set composed of the optimized prompt information corresponding to each lesson plan structure level.

[0047] Specifically, since the multi-level cognitive assessment results include cognitive assessment results corresponding to each lesson plan structure level, optimization prompts corresponding to each lesson plan structure level can be selected from a pre-constructed set of optimization prompts based on each lesson plan structure level. The cognitive assessment results corresponding to each lesson plan structure level are used as inputs for placeholders in the corresponding optimization prompts to complete the construction of target optimization prompts for each lesson plan structure level. Based on each target optimization prompt, LLM is used to optimize the original process logic description information under the corresponding lesson plan structure level, resulting in a target process analysis file that meets the user's cognitive level and can clearly display the internal analysis logic of a large pre-trained language model.

[0048] The technical solution of this embodiment involves acquiring original process logic description information and target user cognitive profiles, and performing structured parsing on the original process logic description information to determine structured skeleton information. Based on the target user cognitive profiles, a multi-level cognitive adaptability assessment is conducted on the structured skeleton information to determine the multi-level cognitive assessment results. The levels of the multi-level cognitive assessment are determined according to the teaching plan structure levels. Based on the multi-level cognitive assessment results and a pre-constructed set of optimization prompts, the original process logic description information is optimized at different levels to determine the target process analysis file. The set of optimization prompts includes optimization prompts corresponding to each level. By adopting the above technical solution, the internal logical reasoning process of a large pre-trained language model is transformed into an analysis process that is understandable to the user, analogous to a "teaching process." After structuring the original process logic description information of the acquired large-scale pre-trained language model, based on the acquired target user cognitive profile containing user cognitive characteristics and the teaching plan structure hierarchy of the teaching plan used in the "teaching process", the original process logic description information is subjected to multi-level cognitive evaluation and optimization in accordance with the teaching plan structure hierarchy. Finally, a target process analysis document is generated that conforms to user cognitive characteristics, exists in a "teaching plan" format, and can better enable users to understand the internal logical reasoning process of the large-scale pre-trained language model. While ensuring the accuracy of the large model analysis process presented to the outside world, the readability of the large model analysis process presented to the outside world is improved.

[0049] Example 2

[0050] Figure 2 This is a flowchart of a large model analysis process generation method provided in Embodiment 2 of the present invention. The technical solution of this embodiment further optimizes the above-mentioned optional technical solutions. After obtaining the original process logic description information, the overall architecture of the original process logic description information is extracted from the overall level of the lesson plan structure hierarchy. Then, for the knowledge module sequences divided into module levels, model-level architecture extraction, including paragraph parsing and teaching element type labeling, is performed. A logical relationship graph is constructed based on the logical dependencies between the content at each level within the model architecture, jointly determining the structured skeleton information of the original process logic description information under different lesson plan structure hierarchies. Then, based on the target user's cognitive profile, the cognitive adaptability of the structured skeleton information is evaluated at the overall level, module level, and paragraph level under the lesson plan structure hierarchy. The obtained overall cognitive evaluation results, module cognitive evaluation results, and paragraph cognitive evaluation results are used together as multi-level cognitive evaluation results to assess the comprehensibility of the original process logic description information relative to the user's cognitive ability. The cognitive assessment results of each level in the multi-level cognitive assessment are substituted into the corresponding optimization prompts in the pre-constructed set of optimization prompts to complete the construction of target optimization prompts for each level. Based on each target optimization prompt, multi-level optimization is performed on the original process logic description information, so that the resulting optimized original process logic description information can exist in the form of a "lesson plan" that is more in line with the user's cognitive ability. After each optimization, its cognitive adaptability relative to the target user's cognitive profile is reassessed until it meets the target user's cognitive needs. Only then is the final optimized original process logic description information determined as the target process analysis file, which improves the user's comprehensibility of the large model analysis process.

[0051] like Figure 2 As shown in Embodiment 2 of the present invention, a method for generating a large model analysis process specifically includes the following steps:

[0052] S201. Obtain the original process logic description information and the target user cognitive profile.

[0053] S202. Perform multimodal content preprocessing, overall structure boundary identification, and knowledge module segmentation on the original process logic description information to determine the multimodal content information, overall structure tree, and knowledge module sequence.

[0054] In this embodiment, multimodal content preprocessing can be specifically understood as the preprocessing of integrating information belonging to different structural types in the original process logic description information into multimodal content of a unified format. For example, Optical Character Recognition (OCR) technology can be used to extract text information from images into processable text, a table structure recognition algorithm can be used to convert table content into structured data format, and a multimedia positioning algorithm can be used to locate the position and type of non-text elements such as images, videos, and audio. The information obtained after all the above processing is then determined as multimodal content information. For example, multimodal content information can be represented as M_content={text_segments,visual_elements, structural_data}, where text_segments is a set of text segments, visual_elements is a set of visual element descriptions, and structural_data is a set of structured data.

[0055] In this embodiment, overall structural boundary recognition can be understood as the process of extracting and recognizing macroscopic structural features in the original process logic description information. For example, overall structural boundary recognition can be achieved through structural template matching technology. That is, a template containing lesson plan structure recognition methods can be pre-set according to the actual situation, and the overall structural boundary of the original process logic description information can be extracted using this template. For instance, a title level recognition algorithm can be used to identify first-level and second-level headings based on features such as font size, position information, and formatting marks; a document segmentation algorithm can be used to identify chapter boundaries based on features such as paragraph separators, line breaks, and semantic jumps in content. Further, after the above recognition, a tree structure can be constructed based on the recognition results as the overall structure tree of the original process logic description information. For example, the overall structure tree can be represented as Structure_tree={(level_i, title_i, content_range_i)}, where level_i represents the level depth of the i-th structural unit, title_i represents the title content, and content_range_i represents the content range.

[0056] In this embodiment, knowledge module segmentation can be understood as a segmentation process that divides the original process logic description information into modules based on different knowledge points, based on the possible division methods of different content modules in the lesson plan shown in the example above, and based on the recognition of the text content in the original process logic description information. A knowledge module sequence can be understood as a sequence composed of content modules corresponding to different knowledge points after segmentation. For example, the segmentation and recognition of different knowledge module content can be achieved through a Topic Change Detection Algorithm. This algorithm can use models such as Bidirectional Encoder Representations from Transformers to convert text content into a semantic vector sequence, identify topic transition points by calculating the semantic similarity changes of adjacent text blocks, detect concept switching boundaries using knowledge graph matching technology, and can also be combined with keyword-based recognition algorithms to identify module delimiters such as "next," "on the other hand," and "the second question," thereby achieving the segmentation of knowledge modules to which different content in the original process logic description information belongs, and combining the segmented knowledge modules to form a knowledge module sequence. For example, a sequence of knowledge modules can be represented as Modules={M1, M2, ...,Mn}, where Mi=(module_title_i, content_i, start_pos_i, end_pos_i) represents the title, content, start position, and end position of the i-th knowledge module.

[0057] S203. Perform paragraph analysis and teaching element type labeling on each knowledge module in the knowledge module sequence to determine the paragraph-sentence hierarchical mapping relationship and the functional tags corresponding to each level of content.

[0058] In this embodiment, paragraph parsing can be specifically understood as performing fine-grained decomposition of the content corresponding to each knowledge module in the knowledge module sequence at the paragraph level. It may also include fine-grained parsing of multiple sentences contained within a paragraph. For example, paragraph parsing may include paragraph boundary detection algorithms based on line breaks, indentation, semantic integrity, etc., to identify natural paragraphs within the knowledge module. Each identified natural paragraph can be decomposed into a sentence sequence belonging to each natural paragraph. Then, syntactic analysis algorithms can be used to identify the grammatical structure and component relationships within the sentences, determining the hierarchical mapping relationship between paragraphs and sentences within the paragraph, and defining the obtained information as a paragraph-sentence hierarchical mapping relationship. For example, the paragraph-sentence hierarchical mapping relationship can be represented as Paragraph_structure={(Pi,{Sij})}, where Pi represents the i-th paragraph, and Sij represents the j-th sentence in paragraph Pi.

[0059] In this embodiment, the labeling of teaching element types can be understood as pre-setting multiple teaching element type labels according to the actual situation, and labeling the teaching elements to which the content of the paragraph-sentence level belongs based on a pre-built teaching element recognition model or other algorithms that can be used to identify teaching elements contained in paragraph-level or sentence-level content. For example, conceptual explanation segments can be identified by keywords such as ["defined as", "refers to", "concept"], case description content can be identified by keywords such as ["for example", "likely", "example"], practical links can be identified by keywords such as ["practice", "calculate", "solve"], and summary content can be identified by closing words such as ["in summary", "in conclusion", "summary"]. Each identified level of content can be labeled with a label containing the function of the teaching element. For example, the functional label labeled for the i-th level of content can be represented as Label_i∈{concept,example, practice, summary, transition}.

[0060] It is understood that the hierarchical content mentioned above may not only refer to the content contained in a single sentence, but may also include the content contained in an incomplete paragraph, a complete paragraph, or multiple paragraphs composed of multiple sentences. This embodiment of the invention does not impose any limitations on this.

[0061] S204. Based on the functional tags corresponding to each level of content, determine the logical dependencies between each level of content, and construct a logical relationship graph corresponding to the original process logical description information based on each logical dependency.

[0062] Specifically, since there are logical dependencies between the teaching elements corresponding to different functional tags, after labeling the content at each level, a pre-built causal relationship identification algorithm can be used to determine the causal logical chain between different levels of content based on the causal relationships between the functional tags. Furthermore, a sequence analysis algorithm can be used to identify the pre-dependencies of the knowledge points corresponding to each level of content. Simultaneously, semantic association analysis technology can be used to determine the semantic similarity and association strength between concepts contained in different levels of content. By integrating the above causal relationships, pre-dependencies of knowledge points, semantic similarity between concepts, and association strength, a logical relationship graph corresponding to the original process logical description information can be constructed based on the logical relationships contained therein. For example, the logical relationship graph can be represented as Logic_graph=(Nodes, Edges), where Nodes is the set of content nodes, Edges is the set of logical relationship edges, and each edge e(ni, nj, type, weight) represents the logical relationship type and strength weight from node ni to nj.

[0063] S205. The multimodal content information, the overall tree structure, the knowledge module sequence, the paragraph-sentence hierarchical mapping relationship, the functional labels and logical relationship graphs corresponding to each level of content are determined as the structured skeleton information.

[0064] Following the example above, the structured skeleton information can be represented as Skeleton_structure={M_content,Structure_tree,Modules, Paragraph_structure,[ ... Label_i ... ], Logic_graph}.

[0065] It is understood that the content contained in the structured skeleton information can be added according to actual needs, as long as it can be calculated from the original process logic description information using the corresponding structure extraction algorithm. This embodiment of the invention does not impose any restrictions on this.

[0066] S206. Based on the target user's cognitive profile, conduct an overall cognitive adaptability assessment of the structured skeleton information and determine the overall cognitive assessment result.

[0067] Specifically, based on the cognitive abilities and learning preferences of the target users contained in the target user cognitive profile, the macro-structural features obtained by analyzing the structured skeleton information at the overall level are evaluated in relation to cognitive adaptability, and the evaluation results are integrated as the overall cognitive evaluation results.

[0068] Optionally, the macroscopic structural features obtained from the overall analysis may include features obtained from analyzing whether the structured skeleton information is logically consistent at the overall level, features obtained from analyzing whether the structured skeleton information conforms to the logical progression in the overall explanation process, features obtained from analyzing whether there are logical breaks, missing causal relationships, or unclosed reasoning chains in the overall logical development of the structured skeleton information, and features obtained from analyzing whether the structured skeleton information conforms to known effective communication models in terms of content organization and expression.

[0069] Optionally, based on the target user's cognitive profile, a holistic cognitive adaptability assessment can be performed on the structured skeleton information to determine the overall cognitive assessment result. This can be achieved in the following ways:

[0070] 1) Based on the paragraph-sentence hierarchical mapping relationship in the structured skeleton information, the functional tags corresponding to each level of content, and the cognitive profile of the target user, evaluate the logical consistency between adjacent paragraphs in the structured skeleton information and determine the logical consistency cognitive matching evaluation result.

[0071] In this embodiment, logical consistency assessment can be specifically understood as an assessment method that detects the logical consistency between adjacent paragraphs in the structured skeleton information and evaluates whether the detected logical consistency conforms to the target user's understanding pattern and cognitive needs in combination with the target user's cognitive profile.

[0072] Specifically, since the logical consistency assessment of structured skeleton information mainly involves the adjacent paragraphs in the structured skeleton information, as well as the functional tags corresponding to the hierarchical content contained in the paragraph and sentence levels to realize the logic between adjacent paragraphs, the paragraph-sentence hierarchical mapping relationship in the structured skeleton information, the functional tags corresponding to each level of content, and the target user cognitive profile can be used as the input for this logical consistency assessment, and the output object is the logical consistency cognitive matching assessment result.

[0073] Following the example above, we can first extract adjacent paragraph pairs Pair={Pi, Pi+1} in natural order based on Paragraph_structure, and further decompose each paragraph into sentence sequences to form sentence input pairs Pair_sentences={(Sij),(Si+1,k)} that can be used as input for logical judgment. Then, we can calculate the semantic cosine similarity sim_global between paragraphs based on a pre-trained text representation model (such as the BERT model or other multilingual embedding models) as a quantitative indicator of semantic coherence. If sim_global < threshold θ1, we can mark "weak topic coherence" in the logical consistency evaluation result. We can also extract and align key entities in adjacent paragraphs using methods such as named entity recognition and homonymy resolution, comparing their attributes (such as numerical values, time, identity, causal relationships) to see if there are direct conflicts. If a conflict is detected, we can mark "entity conflict" in the logical consistency evaluation result and attach conflict evidence. Finally, we can use a natural language inference model, taking paragraph Pi as a premise and paragraph Pi+1 as a basis... Assuming a certain assumption, three probabilities P(ENTAIL), P(CONTRA), and P(NEUTRAL) are calculated, and a logical consistency judgment value logic_score = P(ENTAIL) - P(CONTRA) is generated. If logic_score < threshold θ2, then "potential logical contradiction exists" can be marked in the logical consistency evaluation result. Furthermore, the logical relationship graph Logic_graph in the structured skeleton information and the discourse relationship analysis algorithm can be used to identify the discourse relationship type between paragraph pairs (such as continuation, elaboration, contrast, cause-result, etc.). If contrast is detected and the semantic similarity is high, then "contradictory inconsistency" can be marked in the logical consistency evaluation result. If continuation is detected and the semantic similarity is low, then "insufficient connection" can be marked in the logical consistency evaluation result. Then, the obtained logical consistency assessment results are compared with the indicators such as logical_reasoning and abstract_thinking in User_cognitive_portrait. When logical consistency is insufficient but the user's logical reasoning ability is strong, the user's cognitive adaptability can be marked as "understandable but requires additional cognitive load"; when logical consistency is insufficient and the user's logical reasoning ability is weak, the user's cognitive adaptability can be marked as "high-risk cognitive breakpoint".Furthermore, by combining the above logical consistency evaluation results with user cognitive adaptability, we can obtain the logical consistency cognitive matching evaluation results. The logical consistency cognitive matching evaluation results can be represented as Consistency_evaluation={(Pair_id,label, evidence)}, where Pair_id is the unique identifier of Pair_sentences, consistency_score ∈[0,1], label ∈ {“entity conflict”, “oppositional inconsistency”, “insufficient acceptance”, “high-risk cognitive breakpoint”...}, and evidence is the corresponding text fragment, which serves as interpretable evidence.

[0074] 2) Based on the knowledge module sequence in the structured skeleton information, the functional tags corresponding to each level of content, and the cognitive profile of the target user, evaluate the logical progression of the explanation process corresponding to the structured skeleton information, and determine the matching result between progression and cognitive load.

[0075] In this embodiment, the logical progression rule evaluation can be specifically understood as an evaluation method for whether the explanation process conforms to the logical progression rule and whether it meets the cognitive load threshold of the target user when using structured skeleton information as the basis for explanation, and whether the progressive nature of the corresponding knowledge points will cause cognitive overload or breakage of the target user.

[0076] Specifically, since the evaluation of the logical progression of structured skeleton information mainly involves different knowledge modules and the logical relationships between them, the sequence of knowledge modules in the structured skeleton information, the functional tags corresponding to each level of content, and the cognitive profile of the target user can be used as inputs for this logical progression evaluation. The output is the result of matching progression with cognitive load.

[0077] Following the example above, the input information for evaluating the logical progression pattern is the Logic_graph and Modules in Skeleton_structure, as well as User_cognitive_portrait. The output is the Progression_evaluation result, which matches the progression with the cognitive load. First, based on the logical dependency edges in the Logic_graph, the directed chain Progression_chain={c1→c2→…→cn} from basic concepts to advanced content can be automatically identified, and the knowledge module and paragraph position of each node can be marked. If there are isolated nodes or incomplete paths in the logical chain, they can be marked as "missing progression chain" in the logical chain integrity evaluation result. When there is no "missing progression chain" problem, the concept complexity evaluation model trained with text and complexity as training samples is called to quantify the complexity of each node, forming a complexity sequence {level(c1), level(c2),…}, where level(ci)∈[0,1]. Next, calculate the complexity difference between adjacent nodes Δi = level(ci+1) - level(ci). If Δi is greater than a preset threshold, it can be marked as "leapfrog progression" in the complexity gradient rationality assessment results. Furthermore, the above-obtained complexity difference between adjacent nodes can be compared with Cognitive_load_limits in User_cognitive_portrait. When Δi > intrinsic_threshold, it can be marked as "exceeding the intrinsic cognitive tolerance range" in the cognitive load adaptability assessment results; when the overall information density > extraneous_threshold, it can be marked as "excessively high external cognitive load" in the cognitive load adaptability assessment results; when the above progression chain is too short and lacks intermediate bridging nodes, it can be marked as "low germane support" in the cognitive load adaptability assessment results to indicate to the user that it is difficult to establish a stable knowledge structure. Furthermore, the structure of the progressive chain can be compared with the target user's Learning_preferences. If the user's preferences progress gradually (learning_pace=slow) but the actual chain complexity increases rapidly, it can be marked as "too fast" in the learning preference evaluation results. Conversely, if the user's preferences progress rapidly (learning_pace=fast) but the chain growth is slow, it can be marked as "redundant" in the learning preference evaluation results.Furthermore, by combining the above-mentioned logical chain integrity assessment results, complexity gradient rationality assessment results, cognitive load adaptability assessment results, and learning preference assessment results, the target user's progression and cognitive load matching results can be determined, denoted as Progression_evaluation={(Chain_id,label,evidence)}, where Chain_id is the unique identifier of Progression_chain, label∈{progression is reasonable, progression chain is missing, external cognitive load is too high,...}, and evidence is the corresponding text fragment, serving as interpretable evidence.

[0078] 3) Based on the knowledge module sequence in the structured skeleton information, the functional tags corresponding to each level of content, and the cognitive profile of the target user, evaluate the completeness of the logical unfolding process of the structured skeleton information and determine the logical completeness evaluation result.

[0079] In this embodiment, the integrity assessment of the logic unfolding process can be specifically understood as an assessment of whether there are logical breaks, causal gaps, or incomplete reasoning chains in the structured skeleton information during the logic unfolding process.

[0080] Specifically, since the integrity assessment of the logical unfolding process of structured skeleton information mainly involves different knowledge modules and the logical relationships between them, the sequence of knowledge modules in the structured skeleton information, the functional tags corresponding to each level of content, and the target user's cognitive profile can be used as inputs for this logical unfolding process integrity assessment, and the output object is the logical integrity assessment result.

[0081] Following the example above, the input information for logical integrity assessment is the Logic_graph and Modules in Skeleton_structure, as well as User_cognitive_portrait. The output is the logical integrity assessment result, Integrity_evaluation. First, based on the directed edge set in Logic_graph, a full path traversal is performed on the logical chains within and across each knowledge module, forming a logical path set Path_set={path1, path2,…}. If there are isolated starting or ending nodes, or if the path fails to converge to a reasonable conclusion node, it can be marked as "logical break" in the logical integrity assessment result. Then, a causal relationship identification algorithm is called to perform continuity analysis on the causal edges in the path. If a breakpoint occurs where "the result does not correspond to the antecedent" or "the antecedent does not point to the result," it can be marked as "causal missing" in the logical integrity assessment result. Furthermore, each logical path identified above can be aligned with the core concept nodes in the Skeleton_structure. If the coverage rate is lower than the threshold θ3, it indicates that the logical expansion has not fully covered the key concepts, and it can be marked as "incomplete reasoning chain" in the logical integrity assessment results. Further, based on the above-mentioned markings in the logical integrity assessment results, it can be compared with the logical_reasoning in the target user's cognitive profile. If the logical_reasoning is strong, it can be marked as "implicit reasoning exists, requiring additional thought" in the logical integrity assessment results; if the logical_reasoning is weak, it can be marked as "significant cognitive break" in the logical integrity assessment results. Furthermore, by integrating the above-obtained logical integrity assessment results, the logical integrity assessment result for the target user can be determined, denoted as Integrity_evaluation ={(Path_id, label, evidence)}, where Path_id is a unique identifier for the logical path, label ∈ {logical integrity, logical break, causal deficiency, incomplete reasoning chain, significant cognitive break,...}, and evidence is the corresponding text fragment or logical relationship subgraph, serving as interpretable evidence.

[0082] 4) Based on the knowledge module sequence, paragraph-sentence hierarchical mapping relationship, functional tags corresponding to each level of content in the structured skeleton information, and the cognitive profile of the target user, evaluate the matching degree between the expression structure pattern of the structured skeleton information and the user's comprehension ability, and determine the evaluation result of the expression structure pattern.

[0083] In this embodiment, the expression structure pattern assessment can be specifically understood as an assessment method that judges whether the content organization and expression of a "lesson plan" composed of structured skeleton information conforms to a known efficient communication model pattern, thereby determining whether the expression structure helps users understand and remember.

[0084] Specifically, since the evaluation of the expression structure pattern of structured skeleton information mainly involves different knowledge modules, paragraph hierarchical structure and logical relationships between different levels, the knowledge module sequence, paragraph-sentence hierarchical mapping relationship, functional tags corresponding to each level of content, and target user cognitive profile in the structured skeleton information can be used as input for this expression structure pattern evaluation, and the output object is the expression structure pattern evaluation result.

[0085] Following the example above, the input information for expression structure pattern evaluation is Modules, Paragraph_structure, and Label_i from Skeleton_structure, and the output is the expression structure pattern evaluation result, Expression_evaluation. First, a pre-built expression template library, Mode_template_set, can be constructed based on various existing efficient communication models. This pre-built expression template library can include PREP (Point-Reason-Example-Point), SCQA (Situation-Complication-Question-Answer), STAR (Situation-Task-Action-Result), FIRE (Fact-Insight-Recommendation-Example), RIDE (Role-Intention-Details-Effect), GROW (Goal-Reality-Options-Will), FFC (Fact-Feeling-Call), and a 30-second elevator pitch pattern, etc. Each pattern is stored as a serialized template, Mode_template={Step1,Step2,…,Stepn}, and is equipped with keywords, sentence structure features, and logical relationship constraints. Then, based on the teaching element annotations in Label_i and Paragraph_structure, the functional label sequence {L1, L2, ..., Lm} of the paragraph is extracted. This sequence is then compared with the standard sequence of Mode_template through sequence matching and similarity calculation. If the similarity exceeds the threshold θ4, the paragraph or knowledge module is considered to conform to the corresponding expression structure pattern. If no pattern with similarity exceeding the threshold θ4 exists, the expression structure pattern evaluation result is marked as "structure pattern missing". Furthermore, the patterns in the aforementioned pre-built expression template library may only be partially implemented by the structured skeleton information, such as a missing Example step in PREP or an unspecified Question in SCQA. In such cases, the expression structure pattern evaluation result can be marked as "pattern incomplete," and the missing part can be output as evidence. Furthermore, by combining Learning_preferences in User_cognitive_portrait, the target user's preferences can be matched with the successfully matched patterns of the structured skeleton information. If the user prefers "structured case-oriented" and a STAR pattern match is detected, the expression structure pattern evaluation result can be marked as "high fit"; if the user prefers "quick conclusion-oriented" and the lesson plan uses the lengthy SCQA pattern, the expression structure pattern evaluation result can be marked as "inefficient expression".Furthermore, by integrating the above expression structure pattern evaluation results, we can obtain the expression structure pattern evaluation result for the target user, denoted as Expression_evaluation = {(Module_id, label, evidence)}, where Module_id is the unique identifier of the knowledge module, label ∈ {match PREP, match SCQA, match STAR, incomplete pattern, missing pattern, inefficient expression,...}, and evidence is the corresponding text fragment or label sequence, serving as interpretable evidence.

[0086] 5) The results of the logical consistency cognitive matching assessment, the results of the progression and cognitive load matching assessment, the results of the logical integrity assessment, and the results of the expression structure pattern assessment are determined as the overall cognitive assessment results.

[0087] It is understood that the overall cognitive assessment results may include one or more of the following: logical consistency cognitive matching assessment results, progression and cognitive load matching results, logical integrity assessment results, and expression structure pattern assessment results. This embodiment of the invention does not impose any limitations on this.

[0088] S207. Based on the target user's cognitive profile, conduct a module-level cognitive adaptability assessment of the structured skeleton information and determine the module cognitive assessment results.

[0089] Specifically, based on the cognitive abilities and learning preferences of the target users contained in the target user cognitive profile, the structured skeleton information is evaluated in multiple dimensions and related to cognitive adaptability in the knowledge module dimension. Furthermore, the problem points that may cause cognitive difficulties for the target users at the knowledge module design level are marked, and the obtained evaluation results are integrated as the module cognitive evaluation results.

[0090] Optionally, the evaluation at the knowledge module design level may include: evaluating whether the structured expression pattern of the knowledge module dimensions in the structured skeleton information meets the user's understanding needs; evaluating whether the exercise density involved in the knowledge module is manageable for the target user; evaluating whether the knowledge module adopts mnemonic strategies that conform to the user's memory habits and cognitive characteristics; evaluating whether the knowledge module can provide the target user with diversified and practical application scenarios in terms of application scenario design, so that the user can transfer and practice knowledge; and evaluating whether the application scenarios in the knowledge module can effectively map the knowledge principles for the target user to understand.

[0091] Optionally, based on the target user's cognitive profile, a module-level cognitive adaptability assessment can be performed on the structured skeleton information to determine the module cognitive assessment results. This can be achieved in the following way:

[0092] 1) Based on the knowledge module sequence, paragraph-sentence hierarchical mapping relationship, functional tags corresponding to each level of content in the structured skeleton information, and the cognitive profile of the target user, evaluate the matching degree between the expression structure pattern of each module of the structured skeleton information and the user's comprehension ability, and determine the evaluation result of the module-level expression structure pattern.

[0093] In this embodiment, the module-level expression structure pattern evaluation can be specifically understood as an evaluation method that assesses whether the "lesson plan" composed of structured skeleton information conforms to known efficient teaching and communication models in terms of the organization and expression of content within the knowledge module, thereby determining whether the presentation structure of the knowledge module is conducive to the target user's understanding and mastery. It is understood that this judgment method is similar to the overall hierarchical expression structure pattern evaluation method in S206 4) above.

[0094] Specifically, since the evaluation of the expression structure pattern at the knowledge module level mainly involves knowledge modules, the corresponding paragraph hierarchical structure of knowledge modules, and the logical relationships between different levels, the knowledge module sequence, paragraph-sentence hierarchical mapping relationship, functional tags corresponding to each level of content, and target user cognitive profile in the structured skeleton information can be used as inputs for this module-level expression structure pattern evaluation, and the output object is the module-level expression structure pattern evaluation result.

[0095] Following the example above, the input information for evaluating the module-level expression structure pattern is Modules, Paragraph_structure, and Label_i from Skeleton_structure, and the output is the module-level expression structure pattern evaluation result Module_expression_evaluation. First, based on the Modules partitioning results in Skeleton_structure, module types are identified one by one, such as import, new lesson explanation, classroom exercises, group activities, summary and generalization, and homework assignment. Then, combining Paragraph_structure and Label_i, the discourse structure within the module is functionally labeled, generating a functional label sequence {L1, L2, …, Lm}. Next, by calling the same pre-built expression template library Mode_template_set, sequence matching and semantic similarity calculation are used to compare the module functional label sequence with Mode_template. When the similarity exceeds the threshold θ5, the knowledge module is considered to conform to the corresponding expression structure pattern; otherwise, if no pattern with similarity exceeding the threshold θ5 exists, the module-level expression structure pattern evaluation result is marked as "module expression structure missing". Furthermore, pattern integrity checks can be performed. If a pattern is only partially satisfied (e.g., using the SCQA pattern in the import module but lacking Question phrases; using the PREP pattern in the new lesson explanation but lacking the Example section), the module-level expression structure pattern evaluation result can be marked as "pattern incomplete," and the missing part can be returned as evidence. Further, considering the target user's cognitive profile, when the user's cognitive profile shows strong logical reasoning, patterns with clear logic such as SCQA and STAR can be more recommended; when the user's attention span is short, concise and intuitive expression patterns such as PREP, FFC, and the 30-second elevator pitch pattern can be recommended. If a conflict is detected between the pattern used in the knowledge module and the user's learning preferences, the module-level expression structure pattern evaluation result can be marked as "inefficient expression."Furthermore, by integrating the above module-level expression structure pattern evaluation results, we can obtain the module-level expression structure pattern evaluation results for the target user, denoted as Module_expression_evaluation = {(Module_id, label, evidence)}, where Module_id is a unique identifier for the module, label ∈ {match PREP, match SCQA, match STAR, incomplete pattern, missing pattern, inefficient expression,...}, and evidence is the sequence of functional labels or text fragments for the corresponding module, serving as interpretable evidence.

[0096] 2) Based on the knowledge module sequence, paragraph-sentence hierarchical mapping relationship, functional tags corresponding to each level of content, and target user cognitive profile in the structured skeleton information, conduct practice density assessment on each module of the structured skeleton information and determine the practice density assessment results.

[0097] In this embodiment, the practice density assessment can be specifically understood as an assessment method for whether the quantity, distribution, and difficulty of the practice design in each knowledge module of the structured skeleton information are reasonable, and whether the practice density exceeds the cognitive load of the target user or can form effective training.

[0098] Specifically, since the assessment of practice density within a knowledge module mainly involves the knowledge modules in the structured skeleton information, as well as the practice-related parts of different paragraph hierarchies and content within those hierarchies, the sequence of knowledge modules in the structured skeleton information, the paragraph-sentence hierarchical mapping relationship, the functional tags corresponding to each level of content, and the target user's cognitive profile can be used as inputs for this practice density assessment, and the output object is the practice density assessment result.

[0099] Following the example above, the input information for practice density evaluation is Modules, Paragraph_structure, and Label_i from Skeleton_structure, and the output is the practice density evaluation result Practice_density_evaluation. First, based on Label_i, the content within each knowledge module is filtered to identify content units labeled as "practice," and then categorized into three types based on context: basic practice, extended practice, and immediate feedback, forming a practice set Practice_set={p1,p2,…,pn}. Then, the number of practices within each module is counted and normalized with the total text length and the number of knowledge points in the module to obtain the practice density index: density=|Practice_set| / |Knowledge_points|. If density < threshold θ6, it can be marked as "insufficient practice" in the practice density evaluation result; if density > threshold θ7, it can be marked as "excessive practice" in the practice density evaluation result. Furthermore, the rationality of the distribution of exercises within a knowledge module can be tested based on the aforementioned exercise set to analyze the positional distribution of exercises within the module. If all exercises are found to be concentrated at the end of the module with a lack of process-oriented exercises, this can be marked as "uneven distribution" in the exercise density assessment results. If the exercises within a knowledge module cover multiple stages such as knowledge presentation, key point breakthroughs, and difficult point analysis, this can be marked as "reasonable distribution" in the exercise density assessment results. Further, the density can be compared with indicators such as learning_pace, attention_span, and intrinsic_threshold in User_cognitive_portrait, based on the target user's cognitive profile. When a user's attention_span is short and the number of exercises is excessive, this can be marked as "cognitive overload" in the exercise density assessment results. When a user's domain_familiarity is low and the number of exercises is too small, this can be marked as "undertraining" in the exercise density assessment results. When a user's feedback_sensitivity is high and there is a lack of immediate feedback exercises, this can be marked as "lack of feedback" in the exercise density assessment results.Furthermore, the above-mentioned practice density evaluation results can be integrated to determine the practice density evaluation result for the target user, denoted as Practice_density_evaluation={(Module_id,label,evidence)}, where Module_id is a unique identifier for the module, label∈{reasonable practice, insufficient practice, excessive practice, uneven distribution, cognitive overload, insufficient training, lack of feedback,...}, and evidence is the practice text fragment extracted from the corresponding module, serving as an interpretable basis.

[0100] 3) Based on the knowledge module sequence, paragraph-sentence hierarchical mapping relationship, functional tags corresponding to each level of content, logical relationship map, and target user cognitive profile in the structured skeleton information, conduct mnemonic evaluation on each module of the structured skeleton information and determine the mnemonic evaluation results.

[0101] In this embodiment, mnemonic evaluation can be specifically understood as an evaluation of the "lesson plan" composed of structured skeleton information at the knowledge module level, focusing on whether effective mnemonic techniques are used in the knowledge presentation and consolidation process within the knowledge module, and whether the mnemonic techniques used are in line with the memory habits and cognitive characteristics of the target user.

[0102] Specifically, since the evaluation of mnemonic techniques at the knowledge module level mainly involves knowledge modules and the mnemonic techniques used within paragraph levels corresponding to knowledge modules, and the use of mnemonic techniques is reflected in the content at each level and the logical relationship between each level, the knowledge module sequence, paragraph-sentence level mapping relationship, functional tags corresponding to each level of content, logical relationship map, and target user cognitive profile in the structured skeleton information can be used as input for this mnemonic technique evaluation, and the output object is the mnemonic technique evaluation result.

[0103] Following the example above, the input information for mnemonic evaluation is Modules, Paragraph_structure, Label_i, and Logic_graph from Skeleton_structure, and the output is the mnemonic evaluation result, Mnemonic_evaluation. First, mnemonic feature identification can be performed on the content of the knowledge modules. For example, keyword matching, semantic pattern recognition, and multimodal detection methods can be used to identify potential mnemonic types from the knowledge module content. Examples include: rhyming / song-based mnemonic techniques: identified through rhyming and catchy phrase detection models; initial letter mnemonic techniques: detected by initial letter abbreviation patterns of consecutive words / phrases; image / mind map / location mnemonic techniques: detected content elements related to graphics, charts, and spatial locations; categorization and pyramid mnemonic techniques: detected by logical tree structures or hierarchical organization; and repetition / spaced mnemonic techniques: detected by the repeated occurrence and time interval design of the same knowledge points in the knowledge module. Then, the set of mnemonic types actually used in each knowledge module, M_types={t1,t2,…,tk}, is statistically analyzed, and the matching degree with a preset mnemonic library is calculated. If M_types is empty, the mnemonic evaluation result can be marked as "mnemonic missing"; if M_types only has a single form, the mnemonic evaluation result can be marked as "single mnemonic". Furthermore, the effectiveness of mnemonics in the knowledge module needs to be evaluated. An adaptation analysis of the mnemonics in the knowledge module is performed using learning_preferences and cognitive_load_limits in User_cognitive_portrait: when the user's content_format_preference prefers images, if the module uses image-based or mind-map-based memory techniques, the mnemonic evaluation result is marked as "high adaptation"; when the user's processing_speed is slow but the module uses overly complex location-based memory techniques, the mnemonic evaluation result is marked as "high load"; when the user's feedback_sensitivity is high and the module lacks repetition / spaced practice, the mnemonic evaluation result is marked as "insufficient memory consolidation". Furthermore, the system can also perform an integrity check on the mnemonic devices within the knowledge module. If the mnemonic device is incomplete in its design (such as a mnemonic missing a rhyming ending, a mind map missing key branches, or insufficient repetition), the evaluation result of the mnemonic device will be marked as "incomplete mnemonic device," and the missing part will be output as evidence.Furthermore, by integrating the above mnemonic evaluation results, we can obtain the mnemonic evaluation result for the target user, denoted as Mnemonic_evaluation={(Module_id,label,evidence)}, where Module_id is a unique identifier for the module, label∈{mnemonic reasonable, mnemonic missing, mnemonic singular, mnemonic incomplete, high fit, high load, insufficient memory consolidation,...}, and evidence is the mnemonic text fragment or graphic element in the corresponding module, serving as an interpretable basis.

[0104] 4) Based on the knowledge module sequence, paragraph-sentence hierarchical mapping relationship, functional tags corresponding to each level of content, logical relationship graph, and target user cognitive profile in the structured skeleton information, conduct an application scenario coverage breadth assessment for each module of the structured skeleton information and determine the application scenario coverage breadth assessment results.

[0105] In this embodiment, the application scenario coverage assessment can be specifically understood as an evaluation method that assesses the breadth of coverage of the content contained in each knowledge module of the structured skeleton information in knowledge application scenarios, and whether it can provide target users with diverse and practical application scenarios to support knowledge transfer and practice.

[0106] Specifically, since the assessment of the breadth of application scenario coverage within a knowledge module mainly involves the knowledge module itself, as well as the application scenarios corresponding to the content contained in the multiple paragraph levels of the knowledge module, and also involves the logical relationships between different levels of content, the knowledge module sequence, paragraph-sentence level mapping relationship, functional tags corresponding to each level of content, logical relationship graph, and target user cognitive profile in the structured skeleton information can be used as inputs for this assessment of the breadth of application scenario coverage, and the output object is the assessment result of the breadth of application scenario coverage.

[0107] Following the example above, the input information for evaluating the breadth of application scenario coverage is Modules, Paragraph_structure, Label_i, and Logic_graph from Skeleton_structure. The output is the evaluation result Scenario_coverage_evaluation. First, application scenarios within each knowledge module in the knowledge module sequence can be extracted. For example, based on functional fragments such as example, practice, and task in Label_i, the knowledge application contexts involved within the knowledge module can be identified, forming a scenario set Scenario_set={s1,s2,…,sn}. Each scenario is then labeled with its application domain (e.g., daily life, interdisciplinary applications, social practice, professional skills, cross-cultural communication, etc.). Next, scenario diversity analysis is performed, counting the number of domain categories |Categories| in Scenario_set, and comparing it with a pre-built application scenario coverage standard library (including common intra-disciplinary applications, interdisciplinary applications, real-life applications, future career applications, etc.). When |Categories| < threshold θ8, it can be marked as "insufficient scenario coverage" in the application scenario coverage breadth assessment results; when the identified scenarios are concentrated in a single category, it can be marked as "single scenario" in the application scenario coverage breadth assessment results. Furthermore, scenario representativeness can be detected through semantic similarity analysis, comparing the identified scenarios with typical scenarios in the standard knowledge application base. When the matching degree is higher than a preset threshold, it can be marked as "strongly representative" in the application scenario coverage breadth assessment results; when the matching degree is lower than a preset threshold or the scenario is severely virtualized, it can be marked as "insufficiently representative" in the application scenario coverage breadth assessment results. Furthermore, it can be combined with the target user's cognitive profile. When the target user's domain_familiarity is low, more life-like contextual support for understanding is needed; when the target user's education_level is high and abstract_thinking is strong, interdisciplinary or abstract scenarios are more suitable. If the actual scenario settings within the knowledge module deviate significantly from the above user cognitive needs, it can be marked as "low fit" in the application scenario coverage breadth assessment results.Furthermore, the application scenario coverage breadth evaluation results obtained above can be integrated to determine the application scenario coverage breadth evaluation results for the target user, denoted as Scenario_coverage_evaluation={(Module_id,label,evidence)}, where Module_id is a unique identifier for the module, label∈{reasonable coverage, insufficient scenario coverage, single scenario, insufficient representativeness, low adaptability,...}, and evidence is the application scenario text fragment in the corresponding module, serving as an interpretable basis.

[0108] 5) Based on the knowledge module sequence, paragraph-sentence hierarchical mapping relationship, functional tags corresponding to each level of content, logical relationship map, and target user cognitive profile in the structured skeleton information, evaluate the scene principle mapping relationship of each module of the structured skeleton information and determine the scene principle mapping evaluation result.

[0109] In this embodiment, the scene principle mapping relationship evaluation can be specifically understood as whether the application scenarios corresponding to each knowledge module in the structured skeleton information can be effectively mapped to the corresponding knowledge principles, evaluating the degree of correlation and coverage ratio between the scene and the knowledge points, and thus judging whether there is a scene disconnect or principle missing evaluation method in the knowledge application.

[0110] Specifically, since the evaluation of the mapping relationship between application scenarios and knowledge principles within a knowledge module mainly involves the knowledge module itself, as well as the application scenarios corresponding to the content contained in the paragraphs at multiple levels of the knowledge module, and also involves the logical relationships between different levels of content, the knowledge module sequence, paragraph-sentence level mapping relationship, functional tags corresponding to each level of content, logical relationship graph, and target user cognitive profile in the structured skeleton information can be used as the input for this scenario principle mapping relationship evaluation, and the output object is the scenario principle mapping evaluation result.

[0111] Following the example above, the input information for the scenario principle mapping evaluation is the Modules, Paragraph_structure, Label_i, and Logic_graph in Skeleton_structure. Based on this, the scenario set Scenario_set={s1,s2,…,sn} in step 4) above can be determined, and the output is the scenario principle mapping evaluation result Mapping_evaluation. First, based on the annotations such as concept, definition, and theory in Logic_graph and Label_i, the key knowledge point set Principle_set={p1,p2,…,pm} within the module can be extracted, and its concepts can be standardized using a knowledge graph. Then, scenario-principle alignment is performed. For each scenario si∈Scenario_set and each principle pj∈Principle_set, the semantic relevance score(si,pj) is calculated. Methods include semantic similarity calculation based on embedding and explicit matching based on keywords. When score(si,pj) ≥ threshold θ9, a mapping relationship e(si,pj) can be established, thereby generating a scenario-principle mapping graph Mapping_graph=(Scenario_set,Principle_set,E), where E represents the set of mapping edges. Further, the scenario-principle mapping rate can be calculated based on this, expressed as: mapping_rate=|{pj∈Principle_set|∃si,e(si,pj)∈E}| / |Principle_set|. If mapping_rate ≥ threshold θ10, the scenario-principle mapping of this knowledge module is considered sufficient, and it can be marked as "sufficient mapping" in the scenario-principle mapping evaluation results. If mapping_rate < threshold θ10 and a large number of principles are not covered by the corresponding scenarios, it can be marked as "insufficient mapping" in the scenario-principle mapping evaluation results. Optionally, it is also possible to check whether the scenario only maps to low-level knowledge points (such as definition-type knowledge points) and does not cover high-level knowledge points (such as principle applications and reasoning chains). If such imbalances exist, they can be marked as "mapping imbalance" in the scenario principle mapping evaluation results. If some scenarios cannot be aligned with any principle at all, they can be marked as "scenario disconnect" in the scenario principle mapping evaluation results. Furthermore, when users have strong abstract thinking and logical reasoning abilities, the requirement for mapping completeness is higher; if users have low domain familiarity, they rely more on the intuitive scenario corresponding to the basic principle; therefore, if the actual mapping deviates significantly from the target user's cognitive profile, it can be marked as "low fit" in the scenario principle mapping evaluation results.Furthermore, the above-obtained scene principle mapping evaluation results can be integrated to determine the scene principle mapping evaluation result for the target user, denoted as Mapping_evaluation={(Module_id,label,evidence)}, where Module_id is a unique identifier for the module, label∈{sufficient mapping, insufficient mapping, unbalanced mapping, scene disconnection, low adaptation}, and evidence is the corresponding scene text and principle fragment, serving as an interpretable basis.

[0112] 6) The evaluation results of module-level expression structure pattern, practice density, mnemonic techniques, application scenario coverage, and scenario principle mapping are determined as module cognition evaluation results.

[0113] It is understood that the module cognition assessment results may include one or more of the following: module-level expression structure pattern assessment results, practice density assessment results, mnemonic assessment results, application scenario coverage breadth assessment results, and scenario principle mapping assessment results. This embodiment of the invention does not impose any limitations on this.

[0114] S208. Based on the target user's cognitive profile, conduct a paragraph-level cognitive adaptability assessment of the structured skeleton information and determine the paragraph cognitive assessment results.

[0115] Specifically, based on the cognitive abilities and learning preferences of the target users contained in the target user cognitive profile, the structured skeleton information is designed with sentence expression, cognitive load, and knowledge memory in each paragraph and compared with the user's cognitive abilities and learning preferences to obtain paragraph cognitive assessment results that can reflect potential cognitive difficulties at the paragraph level.

[0116] Optionally, multi-dimensional evaluation at the paragraph level may include an assessment of whether the descriptive techniques used in the sentences of the paragraph can meet the user's understanding needs, whether the examples and situational descriptions used in the paragraph are close to the target user's daily life experience, whether the cases provided in the paragraph have sufficient diversity to support the user's understanding and application from different perspectives, and whether the mnemonic techniques used in the paragraph can help the target user's memory and understanding in the process of knowledge presentation and memory consolidation.

[0117] Optionally, based on the target user's cognitive profile, a paragraph-level cognitive adaptability assessment is performed on the structured skeleton information to determine the paragraph cognitive assessment results. This can be achieved in the following ways:

[0118] 1) Based on the paragraph-sentence hierarchy mapping relationship in the structured skeleton information, the functional tags corresponding to each level of content, and the target user's cognitive profile, evaluate the sentence expression of each paragraph in the structured skeleton information and determine the sentence expression evaluation results.

[0119] In this embodiment, the statement expression evaluation can be specifically understood as an evaluation method that assesses whether the statements constituting a paragraph use analogies or metaphors to help target users establish a correspondence between abstract concepts and familiar experiences.

[0120] Specifically, since the paragraph-level sentence expression assessment mainly involves the expression assessment of the content contained in each sentence within the paragraph, the paragraph-sentence hierarchical mapping relationship in the structured skeleton information, the functional tags corresponding to each level of content, and the target user cognitive profile can be used as the input for this sentence expression assessment, and the output object is the sentence expression assessment result.

[0121] Following the example above, the input information for sentence expression evaluation is Paragraph_structure, Label_i, and User_cognitive_portrait from Skeleton_structure, and the output is the sentence expression evaluation result Analogy_evaluation. First, analogy / metaphor identification is performed on the content of the paragraph. Based on keywords (such as "like," "similar to," "just like," "as if," "as if") and dependency syntax patterns, analogy / metaphor expressions in the paragraph are detected, generating a candidate set Analogy_set={a1,a2,…,ak}. Optionally, for cross-modal analogies (such as metaphors using images or diagrams), a multimodal recognition module can be invoked to supplement the visual elements with annotations. Further, for each analogy expression in the candidate set, its source domain (Source_domain) and target domain (Target_domain) can be extracted, and a mapping pair (Source_domain->Target_domain) can be established. Further, semantic role annotation and knowledge graph comparison can be used to determine whether there are reasonable similarity attributes (such as structural similarity, functional similarity, causal similarity) between the source and target domains. In the absence of common attributes, the evaluation result of the statement expression can be marked as "analogy not valid". Furthermore, the domain_familiarity and abstract_thinking metrics in User_cognitive_portrait can be combined to evaluate the statement expression. When the target user has a high degree of familiarity with the source domain (e.g., common sense, everyday objects), the analogy is easy to understand, and the statement expression evaluation result can be marked as "high fit". When the target user is unfamiliar with the source domain or finds it too technical (e.g., using an analogy from advanced physics to explain a middle school concept), the statement expression evaluation result can be marked as "low fit". When the target user's abstract_thinking is weak, complex abstract analogies (e.g., philosophical metaphors) can lead to the statement expression evaluation result being marked as "difficult to understand". Furthermore, the completeness of analogical expression in paragraphs can be checked. If the analogy only presents the source domain without clearly mapping the target domain, or if the explanation is incomplete, the evaluation result of the expression can be marked as "incomplete analogy". If there is potential ambiguity in the analogy (such as contradiction between the source domain and the target domain in key attributes), the evaluation result of the expression can be marked as "misleading analogy".Furthermore, by integrating the above sentence expression evaluation results, we can obtain the sentence expression evaluation result for the target user, denoted as Analogy_evaluation={(Paragraph_id,label,evidence)}, where Paragraph_id is the unique identifier of the paragraph, label∈{analogy reasonable, analogy invalid, analogy incomplete, analogy misleading, high fit, low fit, difficult to understand, ...}, and evidence is the corresponding analogy text fragment or image element, serving as the interpretable basis.

[0122] 2) Based on the paragraph-sentence hierarchy mapping relationship in the structured skeleton information, the functional tags corresponding to each level of content, and the cognitive profile of the target user, evaluate the relevance to daily life of each paragraph in the structured skeleton information and determine the relevance to daily life evaluation results.

[0123] In this embodiment, the relevance assessment can be specifically understood as an assessment method that evaluates the affinity and comprehensibility of the content by assessing whether the examples, explanations, and scenarios used in each paragraph of the structured skeleton information are close to the daily life experience of the target user.

[0124] Specifically, since the assessment of relevance to daily life at the paragraph level mainly involves the evaluation of the expression of the content contained in each sentence within the paragraph, the paragraph-sentence hierarchical mapping relationship in the structured skeleton information, the functional tags corresponding to each level of content, and the cognitive profile of the target user can be used as the input for this assessment of relevance to daily life, and the output object is the assessment result of relevance to daily life.

[0125] Following the example above, the input information for the relevance assessment is Paragraph_structure, Label_i, and User_cognitive_portrait from Skeleton_structure, and the output is the relevance assessment result Life_relevance_evaluation. First, based on keyword and entity recognition technology, specific things, situations, or activities involved in the paragraph are extracted, such as everyday elements like "supermarket," "bus," "homework," and "sports competition." These everyday elements are then compared with a pre-built knowledge graph to map them to predefined categories such as "daily life," "school learning," "social events," and "professional fields." Furthermore, based on the categories of the aforementioned life-related elements, a life-related matching degree analysis is performed on each life-related element. The element categories in the paragraph are compared with the domain_familiarity in User_cognitive_portrait: when the paragraph instances mainly come from categories familiar to the target user (such as "family and school" scenarios for primary school students, and "academic research and social practice" scenarios for college students), the life-related matching degree assessment result can be marked as "high relevance"; when the paragraph instances mainly come from fields unfamiliar to the target user or relatively higher in level than their comprehension ability (such as using business case studies for middle school students), the life-related matching degree assessment result can be marked as "low relevance"; if the paragraph does not contain any life-related elements and is only an abstract description, the life-related matching degree assessment result can be marked as "lacking life connection". Furthermore, based on a semantic similarity model, the relevance between knowledge points and everyday elements in the paragraph content can be calculated. If the relevance exceeds a preset threshold, the knowledge point can be considered to be reasonably explained through everyday scenarios (e.g., "electric current is like water flow"), and the relevance assessment result can be marked as "effective analogy." If the relevance is below a preset threshold, the knowledge point can be considered to be far-fetched or without logical support in relation to everyday scenarios, and the relevance assessment result can be marked as "far-fetched analogy." Further, the relevance can be evaluated in conjunction with the target user's learning_preferences. When the target user's content_format_preference favors "everyday examples," paragraphs lacking such elements will be marked as "low fit" in the relevance assessment result. When the user's processing_speed is slow, complex professional scenarios will increase cognitive load, and the corresponding paragraphs will be marked as "high load" in the relevance assessment result.Furthermore, by integrating the above relevance to life assessment results, we can obtain the relevance to life assessment results for the target user, denoted as Life_relevance_evaluation={(Paragraph_id,label,evidence)}, where Paragraph_id is the unique identifier of the paragraph, label∈{high relevance, low relevance, lack of life connection, effective life analogy, far-fetched life analogy, low expression fit, high load}, and evidence is the corresponding text fragment, which serves as the interpretable basis.

[0126] 3) Based on the paragraph-sentence hierarchy mapping relationship in the structured skeleton information, the functional tags corresponding to each level of content, and the cognitive profile of the target user, conduct case diversity evaluation on each paragraph of the structured skeleton information and determine the case diversity evaluation results.

[0127] In this embodiment, case diversity assessment can be specifically understood as an assessment method that evaluates whether the cases provided in each paragraph of the structured skeleton information have sufficient diversity, whether they can support understanding and application from different perspectives, and whether they can meet the cognitive needs and learning interests of the target users.

[0128] Specifically, since the case diversity assessment at the paragraph level mainly involves the expression assessment of the content contained in each sentence within the paragraph, the paragraph-sentence hierarchical mapping relationship in the structured skeleton information, the functional tags corresponding to each level of content, and the target user cognitive profile can be used as the input for this case diversity assessment, and the output object is the case diversity assessment result.

[0129] Following the example above, the input information for case diversity assessment is the Paragraph_structure, Label_i, and User_cognitive_portrait from Skeleton_structure, and the output is the case diversity assessment result Case_diversity_evaluation. First, case units are identified within the paragraph content. Based on the example type content in Label_i, the case set Case_set={c1,c2,…,cn} is extracted from the paragraph, and semantic analysis is used to summarize the cases. Then, case categories are performed. By calling classification algorithms and knowledge graphs, cases are mapped to different categories (such as daily life, intradisciplinary knowledge, interdisciplinary, social practice, history and culture, professional skills, etc.), forming the case category set Case_categories. Furthermore, diversity can be measured for each identified case. By statistical analysis, the coverage quantity |Case_categories| of Case_categories can be obtained, and the semantic similarity between cases can be calculated. If the concentration of case categories in the case set is too high or the similarity between cases is too great, the case diversity assessment result can be marked as "case single". If |Case_categories| ≥ threshold θ11 and the semantic differences are significant, the case diversity assessment result can be marked as "case diverse". Furthermore, the suitability of case selection in a paragraph can be evaluated based on the target user's cognitive profile. When the target user's domain_familiarity is low, cases should cover both everyday and introductory categories. If only abstract or subject-specific cases exist, the case diversity evaluation result can be marked as "low fit." When the target user's abstract_thinking is strong, interdisciplinary or complex cases are more suitable. If such cases are lacking, the case diversity evaluation result can be marked as "insufficient diversity." When the target user's learning_preferences show a preference for "case-driven" content_format_preference, if the number of cases is too small, the case diversity evaluation result can be marked as "case missing." Furthermore, the representativeness of each case in the case set can be tested. If the case diversity is high but there are no typical cases corresponding to the core knowledge points, the case diversity evaluation result can be marked as "insufficient representativeness."Furthermore, by integrating the above case diversity evaluation results, we can obtain the case diversity evaluation results for the target user, denoted as Case_diversity_evaluation = {(Paragraph_id,label, evidence)}, where Paragraph_id is a unique identifier for the paragraph, label ∈ {case diversity, case uniformity, insufficient diversity, missing cases, low fit, insufficient representativeness}, and evidence is the corresponding case text fragment, serving as interpretable evidence.

[0130] 4) Based on the paragraph-sentence hierarchical mapping relationship in the structured skeleton information, the functional tags corresponding to each level of content, and the cognitive profile of the target user, conduct mnemonic evaluation on each paragraph of the structured skeleton information and determine the paragraph-level mnemonic evaluation results.

[0131] In this embodiment, paragraph-level mnemonic evaluation can be specifically understood as an evaluation method that assesses whether the content provided in each paragraph of the structured skeleton information is reasonably embedded in the mnemonic design during the knowledge presentation and memory consolidation process, and whether it helps the target user's memory and understanding.

[0132] Specifically, since the evaluation of mnemonic techniques at the paragraph level mainly involves the evaluation of the expression of the content contained in each sentence within the paragraph, the paragraph-sentence hierarchical mapping relationship in the structured skeleton information, the functional tags corresponding to each level of content, and the cognitive profile of the target user can be used as the input for this paragraph-level mnemonic technique evaluation, and the output object is the paragraph-level mnemonic technique evaluation result.

[0133] Following the example above, the input information for paragraph-level mnemonic evaluation is Paragraph_structure, Label_i, and User_cognitive_portrait from Skeleton_structure, and the output is the paragraph-level mnemonic evaluation result, Mnemonic_evaluation. First, the same method as the module-level mnemonic identification method described above is used to identify the mnemonics contained in the paragraph. That is, potential mnemonic types in the paragraph are identified through keywords, semantic patterns, and structured features, including associative memory, mnemonic rhymes / songs, and initial letter mnemonics. Then, the same method as the module-level mnemonic identification method is used to detect the coverage and completeness of mnemonics. When a paragraph is found to contain no mnemonics, it can be marked "Mnemonic Missing" in the paragraph-level mnemonic evaluation result; when only a single mnemonic appears, it can be marked "Single Mnemonic" in the paragraph-level mnemonic evaluation result; when a mnemonic element is incomplete or ambiguous, it can be marked "Incomplete Mnemonic" or "Ambiguous Mnemonic" in the paragraph-level mnemonic evaluation result. Furthermore, the same approach as the module-level mnemonic identification described above can be used to analyze the rationality and suitability of mnemonic techniques based on the target user's cognitive profile. This involves comparing the mnemonic type with the user's content_format_preference, processing_speed, intrinsic_threshold, and other indicators. When the mnemonic form matches the user's preference, it can be marked as "high suitability" in the paragraph-level mnemonic evaluation results. When complex mnemonics exceed the user's cognitive load, it can be marked as "high load" in the paragraph-level mnemonic evaluation results. When repetition / spaced memory techniques that the user prefers are lacking, it can be marked as "insufficient memory reinforcement" in the paragraph-level mnemonic evaluation results. Furthermore, by integrating the above paragraph-level mnemonic evaluation results, we can obtain the target user's paragraph-level mnemonic evaluation result, denoted as Mnemonic_evaluation = {(Paragraph_id, label, evidence)}, where Paragraph_id is a unique marker for the paragraph, label ∈ {mnemonic reasonable, mnemonic missing, mnemonic singular, mnemonic incomplete, mnemonic ambiguous, high fit, high load, insufficient memory reinforcement}, and evidence is the mnemonic text fragment in the corresponding paragraph, serving as interpretable evidence.

[0134] 5) The results of the sentence expression assessment, the relevance to real life assessment, the case diversity assessment, and the paragraph-level mnemonic assessment are determined as the paragraph cognition assessment results.

[0135] S209. The overall cognitive assessment results, module cognitive assessment results, and paragraph cognitive assessment results are determined as multi-level cognitive assessment results corresponding to the hierarchical structure of the lesson plan.

[0136] S210. Substitute the cognitive assessment results of each level in the multi-level cognitive assessment results into the corresponding optimization prompt information in the pre-constructed optimization prompt information set to determine the target optimization prompt information.

[0137] In this embodiment, the pre-built set of optimization prompts may include overall optimization prompts (Prompt_whole) at the whole level, module optimization prompts (Prompt_module) at the knowledge module level, and paragraph optimization prompts (Prompt_paragraph) at the paragraph level.

[0138] Specifically, by substituting the cognitive assessment results of each level into the placeholder positions of the optimization prompts corresponding to its level in the pre-constructed set of optimization prompts, the complete set of optimization prompts is determined as the target optimization prompts for each level.

[0139] For example, the overall optimization prompts at the holistic level may include assessment deficiencies in the overall structure of the lesson plan (such as logical breaks, insufficient progression, and missing expression patterns); the target user's cognitive profile's requirements for the overall level (such as requirements for logical clarity, pacing control, and preference for expression patterns); and optimization goals (such as "ensuring overall logical coherence," "enhancing progression," and "completing structural patterns"). A simple example template is given below, where the parts in [] correspond to the content in the overall cognitive assessment results mentioned above:

[0140] "You are an education expert and instructional design consultant."

[0141] The following are the evaluation results of the overall level of the original lesson plan:

[0142] [Consistency_evaluation]

[0143] [Progression_evaluation]

[0144] [Integrity_evaluation]

[0145] [Expression_evaluation]

[0146] Target user cognitive profile:

[0147] [User_cognitive_portrait]

[0148] Please keep the teaching theme unchanged:

[0149] 1. Adjust the order or structure of modules to ensure logical coherence and reasonable progression;

[0150] 2. Supplement missing overall logical links or structural patterns (such as PREP, SCQA, STAR);

[0151] 3. Remove duplicate or redundant parts to enhance overall compactness;

[0152] 4. Make the overall framework more in line with users' cognitive habits.

[0153] Output: Optimized overall lesson plan structure (module-level headings and order).

[0154] For example, module optimization prompts at the knowledge module level can be used to ensure that each knowledge module meets user needs in terms of functional design, content expansion, and cognitive adaptability, and satisfies module-level evaluation indicators such as expression structure pattern, practice density, mnemonic techniques, application scenario coverage, and scenario-principle mapping rate. The content may include deficiencies in the module cognitive evaluation results (such as "insufficient practice density," "missing mnemonic techniques," "insufficient scenario coverage," "mapping imbalance," etc.); the target user's cognitive profile's needs at the module level (such as preference for case-driven learning, need for repeated practice, and habit of using visual mnemonic techniques); and optimization goal descriptions (such as "supplementing missing practice questions," "adding real-life scenarios," and "introducing highly adaptable mnemonic techniques"). A simple example template is given below, where the parts filled in [] are the corresponding content from the above module cognitive evaluation results:

[0155] "You are an education expert and instructional design consultant."

[0156] The following are the evaluation results for the lesson plan module level:

[0157] [Module_expression_evaluation]

[0158] [Practice_density_evaluation]

[0159] [Mnemonic_evaluation]

[0160] [Scenario_coverage_evaluation]

[0161] [Mapping_evaluation]

[0162] Target user cognitive profile:

[0163] [User_cognitive_portrait]

[0164] Please optimize each module while keeping the overall framework unchanged:

[0165] 1. Import Module: Add scenario creation or life-related guidance;

[0166] 2. New Lesson Explanation Module: Supplement with appropriate mnemonic techniques and optimize knowledge presentation;

[0167] 3. Classroom Exercise Module: Adjust the quantity and distribution of exercises to cover key knowledge points;

[0168] 4. Group Activity Module: Increase task diversity and interactivity;

[0169] 5. Summary Module: Supplementary review of key knowledge points and summary of methods;

[0170] 6. Task module: Assign tasks that match the user's cognitive load.

[0171] Output: Optimized module structure (optimization description for each module + new content).

[0172] For example, paragraph optimization prompts at the paragraph level are used to optimize the fine-grained content of the generated "lesson plan" at the paragraph level, ensuring that the paragraphs meet the cognitive needs of the target users in terms of sentence expression, logical connection, information load, analogy / metaphor, relevance to daily life, diversity of cases, and mnemonic devices. The content may include deficiencies in the paragraph cognitive assessment results (such as "incomplete analogies," "low relevance to daily life," "single case studies," and "lack of mnemonic devices"); paragraph-level-related features in the user cognitive profile (such as attention_span, processing_speed, abstract_thinking, and content_format_preference); and optimization goals (such as "supplementing relatable cases," "adding diverse metaphors," "introducing concise mnemonic devices," and "reducing sentence complexity"). A simple example template is given below, where the parts filled in [] are the corresponding content from the paragraph cognitive assessment results mentioned above:

[0173] "You are an education expert and instructional design consultant."

[0174] The following are the assessment results for the paragraph levels of the lesson plan:

[0175] [Analogy_evaluation]

[0176] [Life_relevance_evaluation]

[0177] [Case_diversity_evaluation]

[0178] [Mnemonic_evaluation]

[0179] Target user cognitive profile:

[0180] [User_cognitive_portrait]

[0181] Please optimize each paragraph individually while keeping the module framework unchanged:

[0182] 1. Optimize sentence structure and reduce excessively long or complex sentences;

[0183] 2. Add appropriate analogies / metaphors to aid in intuitive understanding;

[0184] 3. Incorporate everyday elements to increase relatability;

[0185] 4. Increase the variety of case studies to support different cognitive perspectives;

[0186] 5. Add mnemonic devices (rhymes, abbreviations, etc.) to key knowledge points.

[0187] Output: Optimized paragraph content (paragraph ID + modified text).

[0188] S211. Optimize the original process logic description information at different levels using the optimization prompts for each target. Return the optimized original process logic description information to perform a multi-level cognitive adaptability assessment. The optimized original process logic description information that passes the assessment is determined as the target process analysis file.

[0189] Specifically, a pre-trained large model for optimizing lesson plan content is invoked. The aforementioned target optimization prompts and original process logic descriptions are used as input to the large model to adjust the macro-structure of the original process logic descriptions, optimize knowledge modules while maintaining the overall structure, and optimize each paragraph individually. It is understood that after any of the above adjustments is completed, the corresponding level of cognitive adaptability assessment can be re-executed. If the re-assessment results still contain deficiencies, the target optimization prompts will be reconstructed based on the re-assessment results, and the adjusted process logic descriptions will be optimized again. In other methods, after all the above adjustments are completed, all multi-level cognitive adaptability assessments can be re-executed, and the decision to continue optimization will depend on whether deficiencies are found in the re-assessment results. The optimized original process logic descriptions after all levels have been optimized are then used as the target process analysis file.

[0190] For example, the overall hierarchical optimization result, which uses Prompt_whole and the original process logic description information as input to the large model, can be denoted as Draft_whole_optimized. Draft_whole_optimized can then replace the original process logic description information and be processed again in steps S202-S209. If all evaluation results meet the standards, then the optimization for the knowledge module level can proceed. If any evaluation result does not meet the standards, then the processing in steps S210 and thereafter can continue until the standards are met.

[0191] Following the example above, when optimizing at the knowledge module level, Prompt_module and Draft_whole_optimized can be used as inputs to the large model, and the result can be recorded as Draft_module_optimized. Draft_module_optimized can then be re-evaluated in S202-S209 related to the knowledge module. If all evaluation results meet the standards, then optimization at the paragraph level can begin; if any evaluation result fails to meet the standards, then the processing in S210 and subsequent steps continues, iteratively optimizing the knowledge module level until the standards are met.

[0192] Following the example above, when optimizing at the paragraph level, Prompt_paragraph and Draft_module_optimized can be used as inputs to the large model, and the result can be recorded as Draft_paragraph_optimized. Draft_paragraph_optimized can then be re-evaluated for paragraph-related aspects in S202-S209. If all evaluation results meet the criteria, the output of the large model can be identified as the optimized original process logic description information that has passed the evaluation. If any evaluation result fails to meet the criteria, the processing in S210 and subsequent steps continues, iteratively optimizing the paragraph level until the criteria are met.

[0193] In some embodiments, a lesson plan evaluation module library can be set up to store various evaluation methods used at the overall level, module level, and paragraph level. For example, for the evaluation needs of logical and structural coherence, the lesson plan evaluation module library may include modules for logical consistency evaluation, logical progression evaluation, logical integrity evaluation, and expression structure pattern evaluation; for the evaluation needs of language complexity and cognitive load, the lesson plan evaluation module library may include modules for analogy / metaphor evaluation, relevance to real life evaluation, and case diversity evaluation; for the evaluation needs of knowledge memory design, the lesson plan evaluation module library may include modules for practice density evaluation and mnemonic evaluation; for the evaluation needs of knowledge application indexing, the lesson plan evaluation module library may include modules for scene coverage breadth evaluation and scene principle mapping rate evaluation. It is understood that the aforementioned assessment modules can be used to assess different levels of content within structured skeleton information. For example, the expression structure pattern assessment module can be used at both the overall level and the knowledge module level, and the mnemonic assessment module can be used at both the knowledge module level and the paragraph level. By constructing a lesson plan assessment module library, the applicability and scalability of this invention's embodiments in multi-level structure assessment processes are improved, allowing the indicators of different assessment modules to be flexibly applied to different assessment levels, thus enhancing the universality and scalability of this solution in practical applications. For example, Figure 3 This is an example diagram illustrating a multi-level cognitive adaptability assessment of structured skeleton information using a lesson plan assessment module library, as provided in Embodiment 2 of the present invention. Squares represent the overall level, circles represent the knowledge module level, and triangles represent the paragraph level. It is understood that in different practical application scenarios, more assessment and optimization function modules can be designed, such as modules for visual structure visualization, visual simplicity assessment and optimization, modules for first-appearance knowledge examples and interest-based assessment and optimization for cognitive tasks, and modules for metacognitive monitoring rate, error examples, and correction assessment and optimization for knowledge memorization. This embodiment of the present invention does not impose limitations on these.

[0194] Optionally, in this embodiment of the invention, the lesson plan evaluation module library mentioned above can be replaced with an "evaluation-optimization routing table." This can be understood as designing corresponding prompt templates for the indicators of the corresponding modules in the lesson plan evaluation module and pre-setting them in an evaluation-optimization routing table. This table may contain fields such as ID, indicator category, specific indicator, applicable level, evaluation threshold, evaluation prompt, and optimization prompt. When it is necessary to evaluate or optimize the original process logic description information, the corresponding prompt can be called for execution by the large model. Each prompt can be executed individually or, according to task requirements, merged into a single prompt by level or category for overall execution. The following are some simple examples of evaluation prompts and optimization prompts; this embodiment of the invention does not limit these examples.

[0195] 1) Logical consistency (overall hierarchy)

[0196] Consistency_evaluation_prompt={"You are an education expert. Please evaluate the logical consistency of the overall structure of the following lesson plan. Input: Overall structure of the lesson plan (including module order and content summary). Output: 1. Are there any contradictions or logical breaks? 2. Are there any duplicate or conflicting modules? 3. Tags: {Logical Consistency, Logical Break, Duplication and Contradiction}; 4. Evidence: Text snippets indicating problems."} Consistency_optimization_prompt={"You are an education expert. Please optimize the overall structure of the following lesson plan while keeping the theme unchanged: 1. Adjust the module order to eliminate logical breaks; 2. Delete or merge duplicate content; 3. Ensure overall logical consistency. Output: Optimized overall module structure (module titles + order)."}

[0197] 2) Practice density (module level)

[0198] Practice_density_evaluation_prompt={"Please evaluate the classroom practice section of the following modules: 1. Is the number of practice questions reasonable (covering knowledge points without overloading); 2. Are the types of practice questions diverse (basic, extended, feedback); 3. Tags: {reasonable density, insufficient density, excessive density, single type}; 4. Evidence: Cited practice content."} Practice_density_optimization_prompt={"Please optimize the following classroom practice modules: 1. Add or remove the number of practice questions to achieve a moderate training intensity; 2. Add different types of practice (basic, extended, immediate feedback); 3. Maintain complete knowledge coverage. Output: Optimized classroom practice modules."}

[0199] 3) Mnemonic devices (paragraph level)

[0200] Mnemonic_evaluation_prompt={"Please evaluate whether the following paragraphs contain reasonable mnemonic devices: 1. Do they contain mnemonics, acronyms, categories, or visual cues? 2. Are the mnemonics complete and unambiguous? 3. Do they conform to the cognitive preferences of the target users? 4. Tags: {Reasonable mnemonic device, Missing mnemonic device, Simple mnemonic device, Incomplete mnemonic device, Ambiguous mnemonic device}; 5. Evidence: Cite relevant text snippets."} Mnemonic_optimization_prompt={"Please design suitable mnemonics for the following paragraphs while keeping the knowledge points unchanged: 1. Mnemonics, acronyms, keyword cues, or visual expressions can be used; 2. Ensure the mnemonics are complete, clear, and unambiguous; 3. Conform to the cognitive preferences of the target users. Output: Optimized paragraph content."}

[0201] Optionally, the evaluation and optimization prompt templates can be merged, as shown in the following example:

[0202] {

[0203] You are an education expert and instructional design consultant.

[0204] Based on the following overall hierarchical evaluation results, please optimize the overall framework of the lesson plan.

[0205] enter:

[0206] [Consistency_evaluation]

[0207] [Progression_evaluation]

[0208] [Integrity_evaluation]

[0209] [Expression_evaluation]

[0210] User perception profile:

[0211] [User_cognitive_portrait]

[0212] Task requirements:

[0213] 1. Adjust the order or logical relationship of modules to make the whole flow smoothly;

[0214] 2. Supplement the missing overall logical links or expression structure patterns;

[0215] 3. Redundant parts were removed to enhance compactness;

[0216] 4. Maintain a consistent theme to ensure it is suitable for the target users.

[0217] Output: Optimized overall lesson plan structure (module titles and order).

[0218] }

[0219] The technical solution of this embodiment, after obtaining the original process logic description information, extracts the overall architecture of the original process logic description information from the overall level of the lesson plan structure hierarchy. Then, for the knowledge module sequences divided into module levels, it extracts the model-level architecture, including paragraph analysis and teaching element type labeling. Furthermore, it constructs a logical relationship graph based on the logical dependencies between the content at each level within the model architecture, jointly determining the structured skeleton information of the original process logic description information under different lesson plan structure hierarchies. Then, based on the target user's cognitive profile, it performs cognitive adaptability assessments on the structured skeleton information at the overall level, module level, and paragraph level under the lesson plan structure hierarchy. The obtained overall cognitive assessment results, module cognitive assessment results, and paragraph cognitive assessment results are collectively used as multi-level cognitive assessment results to evaluate the comprehensibility of the original process logic description information relative to the user's cognitive ability. The cognitive assessment results of each level in the multi-level cognitive assessment are substituted into the corresponding optimization prompts in the pre-constructed set of optimization prompts to complete the construction of target optimization prompts for each level. Based on each target optimization prompt, multi-level optimization is performed on the original process logic description information, so that the resulting optimized original process logic description information can exist in the form of a "lesson plan" that is more in line with the user's cognitive ability. After each optimization, its cognitive adaptability relative to the target user's cognitive profile is reassessed until it meets the target user's cognitive needs. Only then is the final optimized original process logic description information determined as the target process analysis file, which improves the user's comprehensibility of the large model analysis process.

[0220] Example 3

[0221] Figure 4 This is a schematic diagram of a large model analysis process generation device provided in Embodiment 3 of the present invention, as shown below. Figure 4 As shown, the large model analysis process generation device includes: an information structuring module 31, a multi-level evaluation module 32, and an analysis document determination module 33.

[0222] The information structuring module 31 is used to acquire the original process logic description information and the target user's cognitive profile, and to perform structured parsing on the original process logic description information to determine the structured skeleton information. The multi-level evaluation module 32 is used to perform multi-level cognitive adaptability evaluation on the structured skeleton information based on the target user's cognitive profile to determine the multi-level cognitive evaluation results. The levels of the multi-level cognitive evaluation are determined according to the teaching plan structure levels. The analysis document determination module 33 is used to optimize the original process logic description information at different levels based on the multi-level cognitive evaluation results and the pre-constructed set of optimization prompts to determine the target process analysis document. The set of optimization prompts includes optimization prompts corresponding to each level.

[0223] The technical solution of this invention transforms the internal logical reasoning process of a large pre-trained language model into a user-understandable analysis process, analogous to a "teaching process." After structuring the original process logic description information of the acquired large pre-trained language model, based on the acquired target user cognitive profile containing user cognitive characteristics and the hierarchical structure of the lesson plan used in the "teaching process," the original process logic description information undergoes multi-level cognitive evaluation and optimization that conforms to the hierarchical structure of the lesson plan. This ultimately generates a target process analysis file that conforms to user cognitive characteristics, existing in a manner similar to a "lesson plan," and better enables users to understand the internal logical reasoning process of the large pre-trained language model. This ensures the accuracy of the large model analysis process presented externally while improving its readability.

[0224] Optionally, the information structuring module 31 is specifically used for: performing multimodal content preprocessing, overall structural boundary identification, and knowledge module segmentation on the original process logic description information to determine the multimodal content information, the overall structure tree, and the knowledge module sequence; performing paragraph parsing and teaching element type labeling on each knowledge module in the knowledge module sequence to determine the paragraph-sentence hierarchical mapping relationship and the functional labels corresponding to each level of content; determining the logical dependencies between each level of content based on the functional labels corresponding to each level of content, and constructing a logical relationship graph corresponding to the original process logic description information based on each logical dependency relationship; and determining the multimodal content information, the overall structure tree, the knowledge module sequence, the paragraph-sentence hierarchical mapping relationship, the functional labels corresponding to each level of content, and the logical relationship graph as structured skeleton information.

[0225] Optional, multi-level evaluation module 32, specifically used for:

[0226] Based on the cognitive profile of the target user, conduct an overall cognitive adaptability assessment of the structured skeleton information and determine the overall cognitive assessment results.

[0227] Based on the target user's cognitive profile, a module-level cognitive adaptability assessment is performed on the structured skeleton information to determine the module cognitive assessment results.

[0228] Based on the cognitive profile of the target user, a paragraph-level cognitive adaptability assessment is performed on the structured skeleton information to determine the paragraph cognitive assessment results.

[0229] The overall cognitive assessment results, module cognitive assessment results, and paragraph cognitive assessment results are defined as multi-level cognitive assessment results corresponding to the hierarchical structure of the lesson plan.

[0230] Optionally, based on the target user's cognitive profile, a holistic cognitive adaptability assessment is performed on the structured skeleton information to determine the overall cognitive assessment results, including:

[0231] Based on the paragraph-sentence hierarchical mapping relationship in the structured skeleton information, the functional tags corresponding to each level of content, and the cognitive profile of the target user, the logical consistency between adjacent paragraphs in the structured skeleton information is evaluated, and the logical consistency cognitive matching evaluation result is determined.

[0232] Based on the knowledge module sequence in the structured skeleton information, the functional tags corresponding to each level of content, and the cognitive profile of the target user, the logical progression of the explanation process corresponding to the structured skeleton information is evaluated to determine the matching result between progression and cognitive load.

[0233] Based on the knowledge module sequence in the structured skeleton information, the functional tags corresponding to each level of content, and the cognitive profile of the target user, the completeness of the logical unfolding process of the structured skeleton information is evaluated, and the logical completeness evaluation result is determined.

[0234] Based on the knowledge module sequence, paragraph-sentence hierarchical mapping relationship, functional tags corresponding to each level of content in the structured skeleton information, and the cognitive profile of the target user, the matching degree between the expression structure pattern of the structured skeleton information and the user's comprehension ability is evaluated, and the evaluation result of the expression structure pattern is determined.

[0235] The results of the logical consistency cognitive matching assessment, the results of the progression and cognitive load matching assessment, the results of the logical integrity assessment, and the results of the expression structure pattern assessment are determined as the overall cognitive assessment results.

[0236] Optionally, based on the target user's cognitive profile, a module-level cognitive adaptability assessment is performed on the structured skeleton information to determine the module cognitive assessment results, including:

[0237] Based on the knowledge module sequence, paragraph-sentence hierarchical mapping relationship, functional tags corresponding to each level of content in the structured skeleton information, and the cognitive profile of the target user, the matching degree between the expression structure pattern of each module of the structured skeleton information and the user's comprehension ability is evaluated, and the evaluation result of the module-level expression structure pattern is determined.

[0238] Based on the knowledge module sequence, paragraph-sentence hierarchical mapping relationship, functional tags corresponding to each level of content, and target user cognitive profile in the structured skeleton information, the practice density of each module of the structured skeleton information is evaluated, and the practice density evaluation results are determined.

[0239] Based on the knowledge module sequence, paragraph-sentence hierarchical mapping relationship, functional tags corresponding to each level of content, logical relationship map, and target user cognitive profile in the structured skeleton information, mnemonic evaluation is performed on each module of the structured skeleton information to determine the mnemonic evaluation results.

[0240] Based on the knowledge module sequence, paragraph-sentence hierarchical mapping relationship, functional tags corresponding to each level of content, logical relationship graph, and target user cognitive profile in the structured skeleton information, the application scenario coverage breadth assessment of each module of the structured skeleton information is carried out to determine the application scenario coverage breadth assessment results.

[0241] Based on the knowledge module sequence, paragraph-sentence hierarchical mapping relationship, functional tags corresponding to each level of content, logical relationship graph, and target user cognitive profile in the structured skeleton information, the scenario principle mapping relationship of each module of the structured skeleton information is evaluated, and the scenario principle mapping evaluation result is determined.

[0242] The evaluation results of module-level expression structure pattern, practice density, mnemonic techniques, application scenario coverage, and scenario principle mapping are determined as module cognition evaluation results.

[0243] Optionally, based on the target user's cognitive profile, a paragraph-level cognitive adaptability assessment is performed on the structured skeleton information to determine the paragraph cognitive assessment results, including:

[0244] Based on the paragraph-sentence hierarchy mapping relationship in the structured skeleton information, the functional tags corresponding to each level of content, and the cognitive profile of the target user, the sentence expression of each paragraph in the structured skeleton information is evaluated, and the sentence expression evaluation results are determined.

[0245] Based on the paragraph-sentence hierarchy mapping relationship in the structured skeleton information, the functional tags corresponding to each level of content, and the cognitive profile of the target user, the relevance of each paragraph in the structured skeleton information to real life is assessed, and the relevance assessment results are determined.

[0246] Based on the paragraph-sentence hierarchical mapping relationship in the structured skeleton information, the functional tags corresponding to each level of content, and the cognitive profile of the target user, a case diversity evaluation is conducted on each paragraph of the structured skeleton information to determine the case diversity evaluation results.

[0247] Based on the paragraph-sentence hierarchical mapping relationship in the structured skeleton information, the functional tags corresponding to each level of content, and the cognitive profile of the target user, the mnemonic evaluation of each paragraph of the structured skeleton information is carried out to determine the paragraph-level mnemonic evaluation results.

[0248] The results of the sentence expression assessment, the relevance to real life assessment, the case diversity assessment, and the paragraph-level mnemonic assessment were determined as the paragraph cognition assessment results.

[0249] Optionally, the analysis file determination module 33 is specifically used for:

[0250] The cognitive assessment results of each level in the multi-level cognitive assessment results are substituted into the corresponding optimization prompts in the pre-constructed set of optimization prompts to determine the target optimization prompts.

[0251] The original process logic description information is optimized at different levels by using the optimization prompts for each target. The optimized original process logic description information is then returned to perform a multi-level cognitive adaptability assessment. The optimized original process logic description information that passes the assessment is then identified as the target process analysis file.

[0252] The large model analysis process generation device provided in the embodiments of the present invention can execute the large model analysis process generation method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0253] Example 4

[0254] Figure 5 This is a schematic diagram of a large model analysis process generation device provided in Embodiment 4 of the present invention. The large model analysis process generation device 40 can be intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, communication base stations, and other suitable computers. The large model analysis process generation device 40 can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0255] like Figure 5As shown, the large model analysis process generation device 40 includes at least one processor 41 and a memory, such as a read-only memory (ROM) 42 and a random access memory (RAM) 43, communicatively connected to the at least one processor 41. The memory stores computer programs executable by the at least one processor. The processor 41 can perform various appropriate actions and processes based on the computer program stored in the ROM 42 or loaded from storage unit 48 into the RAM 43. The RAM 43 can also store various programs and data required for the operation of the large model analysis process generation device 40. The processor 41, ROM 42, and RAM 43 are interconnected via a bus 44. An input / output (I / O) interface 45 is also connected to the bus 44.

[0256] Multiple components in the large model analysis process generation device 40 are connected to the I / O interface 45, including: an input unit 46, such as a keyboard, mouse, etc.; an output unit 47, such as various types of displays, speakers, etc.; a storage unit 48, such as a disk, optical disk, etc.; and a communication unit 49, such as a network card, modem, wireless transceiver, etc. The communication unit 49 allows the large model analysis process generation device 40 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0257] Processor 41 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 41 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 41 performs the various methods and processes described above, such as methods for generating large model analysis processes.

[0258] In some embodiments, the large model analysis process generation method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 48. In some embodiments, part or all of the computer program may be loaded and / or installed onto the large model analysis process generation device 40 via ROM 42 and / or communication unit 49. When the computer program is loaded into RAM 43 and executed by processor 41, one or more steps of the large model analysis process generation method described above may be performed. Alternatively, in other embodiments, processor 41 may be configured to execute the large model analysis process generation method by any other suitable means (e.g., by means of firmware).

[0259] Optionally, embodiments of the present invention also provide a computer program product, including a computer program that, when executed by a processor, implements the large model analysis process generation method provided in any embodiment of the present invention.

[0260] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0261] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0262] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0263] To provide user interaction, the systems and techniques described herein can be implemented on a large model analysis process generation device, which includes: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the large model analysis process generation device. Other types of devices can also be used to provide user interaction; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0264] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0265] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0266] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0267] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for generating a large model analysis process, characterized in that, include: Obtain the original process logic description information and the target user cognitive profile, and perform structured parsing on the original process logic description information to determine the structured skeleton information; Based on the target user's cognitive profile, a multi-level cognitive adaptability assessment is performed on the structured skeleton information to determine the multi-level cognitive assessment results; wherein, the levels of the multi-level cognitive assessment are determined according to the teaching plan structure levels. Based on the multi-level cognitive assessment results and the pre-constructed set of optimization prompts, the original process logic description information is optimized at different levels to determine the target process analysis file; wherein, the set of optimization prompts includes optimization prompts corresponding to each level.

2. The method for generating a large model analysis process according to claim 1, characterized in that, The step of performing structured parsing on the original process logic description information to determine the structured skeleton information includes: The original process logic description information is preprocessed with multimodal content, the overall structure boundary is identified, and knowledge modules are segmented to determine the multimodal content information, the overall structure tree, and the knowledge module sequence. The paragraphs of each knowledge module in the knowledge module sequence are parsed and the teaching element types are labeled to determine the paragraph-sentence hierarchical mapping relationship and the functional tags corresponding to each level of content; Based on the functional tags corresponding to each level of content, determine the logical dependencies between the content at each level, and construct a logical relationship graph corresponding to the original process logical description information based on the logical dependencies. The multimodal content information, the overall structure tree, the knowledge module sequence, the paragraph-sentence hierarchical mapping relationship, the functional tags corresponding to each level of content, and the logical relationship graph are determined as structured skeleton information.

3. The method for generating a large model analysis process according to claim 2, characterized in that, The step of performing a multi-level cognitive adaptability assessment on the structured skeleton information based on the target user's cognitive profile, and determining the multi-level cognitive assessment results, includes: Based on the target user's cognitive profile, a holistic cognitive adaptability assessment is performed on the structured skeleton information to determine the overall cognitive assessment result. Based on the target user's cognitive profile, the structured skeleton information is evaluated for cognitive adaptability at the module level, and the module cognitive evaluation results are determined. Based on the target user's cognitive profile, the structured skeleton information is evaluated for cognitive adaptability at the paragraph level to determine the paragraph cognitive evaluation result. The overall cognitive assessment results, the module cognitive assessment results, and the paragraph cognitive assessment results are determined as multi-level cognitive assessment results corresponding to the hierarchical structure of the lesson plan.

4. The method for generating a large model analysis process according to claim 3, characterized in that, The step of performing an overall cognitive adaptability assessment on the structured skeleton information based on the target user's cognitive profile, and determining the overall cognitive assessment result, includes: Based on the paragraph-sentence hierarchical mapping relationship in the structured skeleton information, the functional tags corresponding to each level of content, and the target user cognitive profile, the logical consistency between adjacent paragraphs in the structured skeleton information is evaluated to determine the logical consistency cognitive matching evaluation result. Based on the knowledge module sequence in the structured skeleton information, the functional tags corresponding to each level of content, and the cognitive profile of the target user, the logical progression of the explanation process corresponding to the structured skeleton information is evaluated to determine the matching result between progression and cognitive load. Based on the knowledge module sequence in the structured skeleton information, the functional tags corresponding to each level of content, and the target user cognitive profile, the completeness of the logical expansion process of the structured skeleton information is evaluated, and the logical completeness evaluation result is determined. Based on the knowledge module sequence, paragraph-sentence hierarchical mapping relationship, functional tags corresponding to each level of content in the structured skeleton information, and the target user cognitive profile, the matching degree between the expression structure pattern of the structured skeleton information and the user's comprehension ability is evaluated, and the evaluation result of the expression structure pattern is determined. The logical consistency cognitive matching assessment result, the progression and cognitive load matching result, the logical integrity assessment result, and the expression structure pattern assessment result are determined as the overall cognitive assessment result.

5. The method for generating a large model analysis process according to claim 3, characterized in that, The step of performing a module-level cognitive adaptability assessment on the structured skeleton information based on the target user's cognitive profile, and determining the module cognitive assessment result, includes: Based on the knowledge module sequence, paragraph-sentence hierarchical mapping relationship, functional tags corresponding to each level of content in the structured skeleton information, and the target user cognitive profile, the matching degree between the expression structure pattern and the user's comprehension ability of each module of the structured skeleton information is evaluated, and the module-level expression structure pattern evaluation result is determined. Based on the knowledge module sequence, paragraph-sentence hierarchical mapping relationship, functional tags corresponding to each level of content in the structured skeleton information, and the target user cognitive profile, the practice density of each module of the structured skeleton information is evaluated, and the practice density evaluation result is determined. Based on the knowledge module sequence, paragraph-sentence hierarchical mapping relationship, functional tags corresponding to each level of content, logical relationship graph, and the target user cognitive profile in the structured skeleton information, mnemonic evaluation is performed on each module of the structured skeleton information to determine the mnemonic evaluation result. Based on the knowledge module sequence, paragraph-sentence hierarchical mapping relationship, functional tags corresponding to each level of content, logical relationship graph, and target user cognitive profile in the structured skeleton information, the application scenario coverage breadth assessment of each module of the structured skeleton information is performed to determine the application scenario coverage breadth assessment result. Based on the knowledge module sequence, paragraph-sentence hierarchical mapping relationship, functional tags corresponding to each level of content, logical relationship graph, and target user cognitive profile in the structured skeleton information, the scene principle mapping relationship of each module of the structured skeleton information is evaluated, and the scene principle mapping evaluation result is determined. The evaluation results of the module-level expression structure pattern, the evaluation results of the practice density, the evaluation results of the mnemonic techniques, the evaluation results of the application scenario coverage breadth, and the evaluation results of the scenario principle mapping are determined as the module cognition evaluation results.

6. The method for generating a large model analysis process according to claim 3, characterized in that, The step of performing a paragraph-level cognitive adaptability assessment on the structured skeleton information based on the target user's cognitive profile, and determining the paragraph cognitive assessment result, includes: Based on the paragraph-sentence hierarchical mapping relationship in the structured skeleton information, the functional tags corresponding to each level of content, and the target user cognitive profile, the sentence expression of each paragraph in the structured skeleton information is evaluated, and the sentence expression evaluation result is determined. Based on the paragraph-sentence hierarchical mapping relationship in the structured skeleton information, the functional tags corresponding to each level of content, and the target user cognitive profile, the relevance to daily life of each paragraph in the structured skeleton information is evaluated, and the relevance to daily life evaluation result is determined. Based on the paragraph-sentence hierarchical mapping relationship in the structured skeleton information, the functional tags corresponding to each level of content, and the target user cognitive profile, a case diversity evaluation is performed on each paragraph of the structured skeleton information to determine the case diversity evaluation results. Based on the paragraph-sentence hierarchical mapping relationship in the structured skeleton information, the functional tags corresponding to each level of content, and the target user's cognitive profile, mnemonic evaluation is performed on each paragraph of the structured skeleton information to determine the paragraph-level mnemonic evaluation result. The evaluation results of the statement expression, the evaluation results of relevance to daily life, the evaluation results of case diversity, and the evaluation results of paragraph-level mnemonic techniques are determined as paragraph cognition evaluation results.

7. The method for generating a large model analysis process according to claim 1, characterized in that, The step involves optimizing the original process logic description information at different levels based on the multi-level cognitive assessment results and the pre-constructed set of optimization prompts, to determine the target process analysis file, including: The cognitive assessment results of each level in the multi-level cognitive assessment results are substituted into the corresponding optimization prompt information in the pre-constructed set of optimization prompt information to determine the target optimization prompt information; The original process logic description information is optimized at different levels using the target optimization prompts. The optimized original process logic description information is then returned to perform the multi-level cognitive adaptability assessment. The optimized original process logic description information that passes the assessment is determined as the target process analysis file.

8. A device for generating a large model analysis process, characterized in that, include: The information structuring module is used to acquire the original process logic description information and the target user cognitive profile, and to perform structured parsing on the original process logic description information to determine the structured skeleton information; The multi-level assessment module is used to perform multi-level cognitive adaptability assessment on the structured skeleton information based on the target user cognitive profile, and determine the multi-level cognitive assessment results; wherein, the levels of the multi-level cognitive assessment are determined according to the teaching plan structure levels. The analysis file determination module is used to optimize the original process logic description information at different levels based on the multi-level cognitive assessment results and the pre-constructed set of optimization prompts, and determine the target process analysis file; wherein, the set of optimization prompts includes optimization prompts corresponding to each level.

9. A device for generating a large model analysis process, characterized in that, include: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, which is then executed by the at least one processor to enable the at least one processor to perform the large model analysis process generation method according to any one of claims 1-7.

10. A storage medium containing computer-executable instructions, characterized in that, The computer-executable instructions, when executed by a computer processor, are used to perform the large model analysis process generation method as described in any one of claims 1-7.

11. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the method for generating a large model analysis process as claimed in any one of claims 1-7.

Images