A method and system for refining information value based on cognitive hierarchy progression
By developing a method and system for extracting information value based on cognitive hierarchy, this approach solves the problem of multi-level extraction that is difficult to achieve in existing technologies. It generates multi-level cognitive results, improves the efficiency and interpretability of text information utilization, and meets the needs of personalized learning and decision support.
Patent Information
- Application Number
- CN202611075530.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-20
- Publication Date
- 2026-11-10
AI Technical Summary
Existing technologies struggle to achieve a progressive refinement from data to information, knowledge, cognition, and then to wisdom. They cannot simultaneously meet the differences in cognitive depth, granularity of understanding, and decision-making needs among different users, and lack hierarchical organization and cognitive mapping, resulting in insufficient interpretability of the results.
This paper presents a method and system for extracting information value based on cognitive hierarchy. Through sentence segmentation, word segmentation, entity recognition, and semantic structure parsing, it generates multi-level outputs of information layer, knowledge layer, cognitive layer, and wisdom layer. Combined with user needs and application scenario constraints, it performs contextualized reasoning to generate multi-level cognitive results.
It enables the extraction and output of multiple cognitive results corresponding to the same input text, improves the efficiency of text information utilization and the interpretability of results, meets the needs of personalized learning services and decision support, and enhances the observability and interpretability of text information.
Smart Images

Figure CN123070280A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of text information processing, specifically to a method and system for extracting information value based on cognitive hierarchy progression. Background Technology
[0002] With the development of technologies such as natural language processing, knowledge graphs, and large language models, text information processing has gradually expanded from traditional keyword retrieval, automatic summarization, and topic classification to areas such as information extraction, semantic understanding, question answering, and content generation. Existing technologies can typically extract facts, generate summaries, or provide question-answering responses from input text, but their outputs mostly remain at a single level, making it difficult to simultaneously generate continuous, multi-level refinements of information, knowledge, cognition, and even wisdom from the same text.
[0003] Currently, the closest existing technologies mainly fall into the following categories: one category is text processing methods based on rules or statistical models, which typically extract explicit information from text through word segmentation, entity recognition, keyword extraction, and template matching; the other category is generative methods based on pre-trained language models or large language models, which can summarize, rewrite, answer questions, and semantically summarize text, but the output content is usually still a single summary, a single question-and-answer result, or a single-dimensional label, lacking a hierarchical output mechanism for different cognitive levels.
[0004] The aforementioned existing technologies have the following shortcomings: First, the text processing results are often geared towards general summaries or single question-and-answer sessions, making it difficult to achieve a progressive extraction from data to information, knowledge, cognition, and then wisdom; Second, the same input text usually corresponds to only one output result, which cannot simultaneously meet the differences in cognitive depth, granularity of understanding, and decision-making needs of different users; Third, existing methods lack hierarchical organization and cognitive mapping of the output results, resulting in insufficient interpretability of the results and making it difficult to support application scenarios such as personalized learning services, thematic analysis, and decision support. Summary of the Invention
[0005] To address the problems existing in the prior art, this invention provides a method and system for extracting information value based on cognitive hierarchy progression.
[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for extracting information value based on cognitive hierarchy, comprising the following steps: S1. Obtain the target text data to be processed, and perform sentence segmentation, word segmentation, entity recognition, topic recognition and semantic structure parsing on the target text data to obtain the basic text representation; S2. Based on the text basic representation, extract fact items, event items, attribute items, relationship items and context items in the text to generate information layer results for representing the original facts and explicit information in the text; S3. Based on the information layer results, perform knowledge aggregation and rule induction, cluster, deduplicate, associate and merge multiple information elements, construct knowledge units, and generate knowledge layer results to represent the rules, methods and experiences extracted from the information. S4. Based on the knowledge layer results, construct a cognitive framework, perform hierarchical organization, conflict resolution, perspective alignment and semantic compression of knowledge units, and generate cognitive layer results to represent the understanding structure, judgment structure or reasoning structure of the target text. S5. Based on the cognitive layer results, combined with application scenario constraints and user demand parameters, contextual reasoning is performed on the cognitive layer results to generate intelligent layer results for outputting suggestions, decision references, action strategies, or personalized learning paths. S6. Return the information layer results, knowledge layer results, cognition layer results and wisdom layer results as multi-level outputs to achieve the extraction and output of multiple cognitive results corresponding to the same input text.
[0007] On the other hand, the present invention also provides an information value extraction system based on the above method, comprising: The text acquisition module retrieves target text data from the user interface or file storage; The text parsing module performs sentence segmentation, word segmentation, entity recognition, topic recognition, and semantic structure analysis on the text to generate basic text representations. The information extraction module extracts facts, events, attributes, relationships, and contextual items from the basic text representation to generate information layer results; The knowledge construction module clusters, deduplicates, associates, and summarizes the information layer results to build knowledge units and generate knowledge layer results. The cognitive organization module performs hierarchical organization, conflict resolution, perspective alignment, and semantic compression of knowledge units to generate cognitive layer results. The intelligent output module combines user demand parameters and application scenario constraints to perform contextualized reasoning on the cognitive layer results and generate intelligent layer results. The multi-layer output module integrates the results from the information layer, knowledge layer, cognition layer, and wisdom layer into a multi-level output and returns it to the user.
[0008] The application scenario constraints include time constraints, resource constraints, risk constraints, and policy constraints; the user requirement parameters include user cognitive level, learning objectives, professional background, application scenario, and output granularity.
[0009] The beneficial effects of this invention are: This invention provides a method and system for extracting information value based on cognitive hierarchy progression. It enables multiple outputs of results at different cognitive levels from any input text, allowing the same input text to correspond to multiple cognitive outcomes. It constructs a hierarchical processing chain from information, knowledge, cognition to wisdom, allowing the value of textual information to continuously increase with cognitive level, ultimately transforming into understanding, judgment, and decision-making abilities. It provides a differentiated output mechanism for users with different cognitive levels, meeting the needs of application scenarios such as personalized learning services, thematic analysis, knowledge services, and decision support. The text processing process is defined as a technical solution with clear input, processing steps, hierarchical mapping rules, and output structure, resulting in observable technical effects in text parsing, information organization, and intelligent output. This invention, without relying on large-scale data, combines existing technologies such as large language models to construct a text information processing method that takes multiple arbitrary texts as input and outputs multi-layered cognitive results. This overcomes the shortcomings of existing technologies in cognitive hierarchy organization and multi-perspective matching capabilities, improving the efficiency of text information utilization and the interpretability of results. Attached Figure Description
[0010] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation
[0011] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 are within the scope of protection of the present invention.
[0012] Please see Figure 1 This invention discloses a method for extracting information value based on cognitive hierarchy, comprising the following steps: Step S1: Text Acquisition and Basic Representation Construction. The target text data to be processed is acquired, and the target text data is segmented into sentences, words, entities, topics, and semantic structures to obtain the basic text representation. The basic text representation includes at least one or more of the following: text vector representation, paragraph structure representation, keyword representation, and discourse logic representation. This provides a unified input for subsequent extraction of information, knowledge, cognition, and wisdom.
[0013] Step S2: Information layer result generation. Based on the aforementioned textual basic representation, factual items, event items, attribute items, relational items, and contextual items in the text are extracted to generate information layer results. Information layer results are used to answer "what the text says," representing the original facts and explicit information in the text; they represent the first layer of refined results from the cognitive hierarchy.
[0014] Step S3: Knowledge Layer Result Generation. Based on the information layer results, knowledge aggregation and rule induction are performed. Multiple information elements are clustered, deduplicated, associated, and merged to construct knowledge units and generate knowledge layer results. Knowledge layer results include at least one or more of the following: concept association, rule summary, method abstraction, and experience induction. These results are used to answer the question "What rules does the text illustrate?" and represent the rules, methods, and experiences extracted from the information.
[0015] Step S4: Cognitive Layer Result Generation. Based on the knowledge layer results, a cognitive framework is constructed. Knowledge units are hierarchically organized, conflict resolved, perspective aligned, and semantically compressed to generate cognitive layer results. The cognitive framework includes one or more of the following: hierarchical framework, causal framework, contrastive framework, temporal framework, and question-answer framework. The cognitive layer results are used to characterize the understanding structure, judgment structure, or reasoning structure of the target text, answering the question "What understanding framework does the text form?" This is the third layer of refinement in the cognitive hierarchy.
[0016] Step S5: Generating the Wisdom Layer Result. Based on the cognitive layer result, combined with application scenario constraints and user requirement parameters, contextualized reasoning is performed on the cognitive layer result to generate the wisdom layer result. User requirement parameters include one or more of the following: user cognitive level, learning objectives, professional background, application scenario, and output granularity. The wisdom layer result is used to output suggestions, decision-making references, action strategies, or personalized learning paths, answering the question "How should one choose or act when faced with this type of problem?" It represents the highest-level refined result of the cognitive layer and is used to guide choices and actions in specific situations.
[0017] Step S6: Constructing Multi-Level Outputs. The information layer results, knowledge layer results, cognition layer results, and wisdom layer results are returned as multi-level outputs to extract multiple cognitive results corresponding to the same input text. Multi-level outputs can take the form of a tree structure, hierarchical list, structured JSON, or combined text, facilitating subsequent retrieval, teaching, analysis, or decision support.
[0018] Specifically, in step S1, the input raw text is parsed in a structured manner to generate a unified basic representation, providing standardized input for the subsequent extraction of information, knowledge, cognition and wisdom layers.
[0019] The system retrieves target text data from the user interface, file storage, database, or network interface. If the text is a collection of multiple documents, it is processed sequentially according to document number or time order, or multiple documents are merged and processed uniformly. The text is encoded in a unified format (e.g., UTF-8) to remove garbled characters and invisible characters.
[0020] Remove noise from the text, including HTML tags, advertisements, headers and footers, and irrelevant symbols. Compress redundant whitespace characters, reducing consecutive spaces to a single space and consecutive line breaks to a single line break. Standardize punctuation in Chinese text, for example, by unifying full-width punctuation to commonly used forms.
[0021] The text is segmented into sentences based on sentence-ending markers such as periods, question marks, and exclamation marks. For long sentences, they can be appropriately segmented based on semantic connectors (such as "because," "therefore," "but," and "moreover") to ensure the semantic integrity of each sentence. The start and end positions of each sentence are recorded to form a sentence list.
[0022] Each sentence is segmented into words to obtain a word sequence. For Chinese text, standard Chinese word segmentation tools can be used. Each word is then tagged with its part of speech (e.g., noun, verb, adjective), which serves as the basis for subsequent entity recognition and semantic analysis.
[0023] Named entity recognition is performed on word sequences to identify entities such as person names, place names, organization names, time periods, numbers, and technical terms. Each entity is tagged with its type and its position in the text is recorded. Entities are then bound to corresponding sentences to form an entity-sentence association structure.
[0024] Based on keyword statistics, topic modeling, or pre-trained language models, identify the keywords and core topics of the text. Extract the keywords for each paragraph and aggregate them into a full-text keyword set. Label the topic category to which each sentence belongs.
[0025] Perform dependency parsing or semantic role labeling on sentences to identify subject-verb-object structures, modification relationships, causal relationships, conditional relationships, etc. Construct semantic relationships between sentences, such as parallel, progressive, adversative, and causal relationships, to form the logical structure of the discourse. Record the logical relationships between paragraphs, such as introduction, body, and conclusion.
[0026] Each sentence and the entire text are input into a pre-trained language model to obtain sentence vector representations and full-text vector representations. These vector representations are then associated with information such as sentences, paragraphs, entities, and topics to form a complete basic text representation.
[0027] In step S1, the system transforms the original text into a structured basic representation, including sentences, words, entities, topics, semantic structures, and vector representations, providing a unified and operable input for the subsequent extraction of information, knowledge, cognition, and wisdom layers.
[0028] In step S2, fact items, event items, attribute items, relationship items, and context items are extracted from the basic text representation to form an information layer result that represents the original facts and explicit information in the text.
[0029] Identify declarative facts from the text, such as "an event occurred at a certain time and place" or "an object has a certain attribute." Record each fact item with elements such as fact type, subject, time, place, and state. Bind the fact items to the corresponding sentences to form fact-sentence associations.
[0030] Identify events in the text, including event type (e.g., meeting, disaster, transaction, policy release), participants, time, location, process, and result. Record the event skeleton for each event item: event type + participants + time + location + process + result. Associate event items with corresponding sentences and paragraphs.
[0031] Extract object attributes from text, such as "the price of a product," "a person's age," and "the size of an organization." Record each attribute item with: object, attribute name, attribute value, and the sentence from which the data originates. Bind the attribute item to the corresponding entity and sentence.
[0032] Identify relationships between entities, such as "A is the founder of B", "C collaborates with D", "E is located in city F", etc. Record the following for each relationship item: subject, relationship type, object, and source sentence. Bind the relationship item to the corresponding entity and sentence.
[0033] Identify contextual information in the text, such as background information, conditions, restrictions, and exceptions. Record the following for each context item: context type, content, scope of application, and source sentence. Bind the context item to the corresponding paragraph entity.
[0034] Integrate fact items, event items, attribute items, relationship items, and context items into a unified information layer result structure. Initially mark duplicate or conflicting information for subsequent processing steps. Output the information layer result as input for step S3.
[0035] In step S2, the system extracts explicit information from the text into structured information items, forming information layer results and answering "what the text says", providing basic factual input for subsequent knowledge, cognition and wisdom layer extraction.
[0036] In step S3, based on the information layer results, knowledge aggregation and rule induction are performed to construct knowledge units, generate knowledge layer results, and represent the rules, methods and experiences extracted from the information.
[0037] Clustering is performed on fact items, event items, attribute items, relationship items, and context items, grouping items with similar semantics or the same theme into one category. Clustering criteria include: topic similarity, entity overlap, and vector distance. This forms several information clusters, each corresponding to a potential knowledge theme.
[0038] Duplicate information within the same cluster is deduplicated, and information items with the same or highly similar semantics are merged. Conflicting information is marked for conflict resolution in subsequent steps. This results in a streamlined information set, which serves as the basis for constructing knowledge units.
[0039] Extract core concepts from the information set to form concept nodes. Record the following for each concept node: concept name, definition, related attributes, and source information. Associate the concept node with its corresponding information items.
[0040] Extract the relationships between concepts from relation items and event items to form relation edges. Record the following for each relation edge: source concept, relation type, target concept, and source information item. Construct a concept-relationship network to form the basic structure of a knowledge unit.
[0041] Summarize multiple information clusters to identify recurring patterns, trends, or regularities. For each regularity, summarize and record: regularity description, applicable conditions, source information cluster, and confidence level. Compile a list of regularity summaries as part of the knowledge layer results.
[0042] Abstract reusable methods or processes from event items and attribute items, such as operation steps, processing flows, and strategies. Record the following for each method abstraction: method name, step sequence, applicable scenario, and source information. This forms a list of method abstractions.
[0043] Summarize empirical conclusions from the information set, such as "a certain result is usually produced under certain conditions" or "a certain practice is more effective." Record each empirical summary: empirical description, applicable conditions, source information, and confidence level. Create a list of empirical summaries.
[0044] The concepts, rules, methods, and experiences are integrated into a unified knowledge layer structure. Knowledge units are numbered and hierarchically labeled to form a knowledge catalog. The knowledge layer results are output as input for step S4.
[0045] In step S3, the system aggregates the information layer results into knowledge units, forming knowledge layer results that answer the question "What pattern does the text illustrate?", thus realizing the extraction of knowledge from information.
[0046] In step S4, a cognitive framework is constructed based on the knowledge layer results. Knowledge units are hierarchically organized, conflict resolved, perspective aligned, and semantically compressed to generate cognitive layer results that represent the understanding structure, judgment structure, or reasoning structure of the text.
[0047] Based on the topic and structure of the knowledge layer results, select an appropriate cognitive framework type, including: Hierarchical framework: used to represent hierarchical relationships between concepts; Causal framework: used to represent causal relationships between events; Contrast frame: used to represent comparisons between different objects or solutions; Sequence frame: used to represent the temporal order of events; Question-answer frame: used to represent questions and their corresponding answers.
[0048] One or more frames can be selected for combination.
[0049] Within a hierarchical framework, knowledge units are organized according to hierarchical relationships to form a hierarchical structure. Each level records: level number, concept set, and relationship type, thus forming a knowledge hierarchy tree.
[0050] Within a causal framework, causal relationships are extracted from event items and pattern summaries to construct causal chains. For each causal chain, the following are recorded: cause event, result event, causal relationship type, and confidence level. This forms a causal graph structure.
[0051] Within the comparative framework, different objects, solutions, or viewpoints are compared to form a comparative structure. For each comparison item, the following information is recorded: the object of comparison, the dimensions of comparison, the points of difference, and the source. This forms a comparison matrix.
[0052] Within a temporal framework, events are arranged chronologically to form a temporal structure. For each point in time, the following are recorded: time, event, and state change. This forms a timeline.
[0053] Within the question-and-answer framework, questions and their corresponding answers are identified from the text, forming question-and-answer pairs. For each question-and-answer pair, the following are recorded: question description, answer content, and source sentence. This results in a question-and-answer list.
[0054] Conflict resolution is performed on contradictory information within knowledge units: the confidence level of each piece of information is calculated; priorities are determined based on factors such as source authority, chronological order, and semantic consistency; low-priority information is corrected or deleted, or marked as "controversial." This results in a knowledge set with better consistency.
[0055] Aligning different perspectives on the same topic: Identifying different perspectives (such as academic, policy, and practical perspectives); grouping related knowledge units according to perspective; forming a multi-perspective knowledge structure. This provides perspective information for subsequent contextualized reasoning in the intelligence layer.
[0056] Semantic compression of knowledge units is performed to form high-level summaries and understanding frameworks: core concepts and key relationships are extracted; compressed into simplified cognitive structures; and a description of the understanding structure is formed. The understanding structures are numbered and labeled to form a cognitive framework diagram.
[0057] The understanding structure, judgment structure, and reasoning structure are integrated into a unified cognitive layer result. The cognitive framework is visualized (e.g., tree diagrams, cause-effect diagrams, comparison tables) for easy subsequent presentation. The cognitive layer result is output as input for step S5.
[0058] Through step S4, the system organizes the knowledge layer results into a cognitive framework, forming cognitive layer results, answering "what understanding framework does the text form", thus realizing the extraction from knowledge to cognition.
[0059] In step S5, based on the cognitive layer results, combined with application scenario constraints and user demand parameters, the wisdom layer results are generated to output suggestions, decision references, action strategies, or personalized learning paths to guide choices and actions in specific situations.
[0060] Obtain user-required parameters from user input, user profiles, or system settings, including: User cognitive level: such as beginner, intermediate, expert; Learning objectives: such as understanding basic concepts, mastering methods, and conducting in-depth analysis; Professional background: such as medicine, law, engineering, education, etc.; Application scenarios: such as learning, research, decision-making, and practice; Output granularity: such as summary, detailed, operational level.
[0061] Obtain application scenario constraints from system configuration or user input, including: Time constraints: such as requiring the task to be completed within a limited time; Resource constraints: such as available tools, data, and manpower; Risk constraints: such as risk control requirements; Policy constraints: such as regulations and policy requirements.
[0062] Based on the cognitive layer results, combined with user needs parameters and application scenario constraints, contextualized reasoning is performed: mapping the cognitive framework to specific situations; adjusting the output depth according to the user's cognitive level; selecting relevant knowledge and methods based on professional background; and generating targeted suggestions or strategies based on the application scenario. This results in contextualized reasoning outcomes.
[0063] Extract suggestions from the contextualized reasoning results, such as "It is recommended to adopt a certain method" or "It is recommended to focus on a certain issue." Record the following for each suggestion: suggestion content, applicable conditions, source of evidence, and confidence level. Create a suggestion list.
[0064] Extract decision-making references from the contextualized reasoning results, such as decision basis, trade-off factors, and risk points. Record each decision reference with: decision problem, basis, trade-off factors, risk point, and source. Create a decision reference list.
[0065] Extract action strategies from the contextualized reasoning results, including the sequence of steps, key operational points, and precautions. Record the following for each action strategy: strategy name, sequence of steps, applicable scenario, and precautions. Create a list of action strategies.
[0066] Based on the user's cognitive level and learning goals, suitable learning content is extracted from the cognitive level results to form a personalized learning path. The learning path is recorded including: learning stage, learning topic, recommended content, learning order, and expected goals. This results in a personalized learning path.
[0067] Integrate suggestions, decision-making references, action strategies, and personalized learning paths into a unified wisdom layer result. Label the wisdom layer result hierarchically and specify the output format (e.g., text, list, step diagram). Output the wisdom layer result as input for step S6.
[0068] Through step S5, the system contextualizes the cognitive layer results to generate wisdom layer results, answering "how to choose or act when faced with this type of problem," thus realizing the refinement from cognition to wisdom and guiding choices and actions in specific situations.
[0069] In step S6, the information layer results, knowledge layer results, cognition layer results, and wisdom layer results are integrated into a multi-level output, realizing the extraction and output of multiple cognitive results corresponding to the same input text.
[0070] The results from the information layer, knowledge layer, cognition layer, and wisdom layer are organized into a hierarchical structure according to the order of cognitive hierarchy. Each layer records: layer name, layer number, and content set. This forms a cognitive hierarchy tree.
[0071] Select the output format based on user needs and application scenarios, including: Tree structure: suitable for displaying hierarchical relationships; Hierarchical list: suitable for reading and printing; Structured JSON: Suitable for program processing and retrieval; Combined text: Suitable for direct display to users.
[0072] Integrate the results from each level into a multi-level output in the selected format. Add a title, number, and brief description to each level. Ensure there are clear hierarchical relationships and jump links (if applicable) between levels.
[0073] Perform consistency checks on multi-level output to ensure: no duplication or conflict in content across levels; clear hierarchical relationships; and output format conforms to user requirements. Correct or mark problematic content.
[0074] Return multi-level output to the user interface, file storage, database, or network interface. Record output logs, including input text, output time, output format, and output size. Complete the method flow.
[0075] Through step S6, the system integrates the four-layer results of information, knowledge, cognition and wisdom into multi-level output, so that the same input text corresponds to multiple cognitive results, meets the needs of multi-perspective cognitive matching and personalized learning services, and improves the efficiency of text information utilization and the interpretability of results.
[0076] The information layer provides explicit information such as facts, events, attributes, relationships, and context; the knowledge layer provides abstract knowledge such as rules, methods, and experiences; the cognitive layer provides understanding structures, judgment structures, and reasoning structures; and the wisdom layer provides suggestions, decision-making references, action strategies, and personalized learning paths. These four layers together constitute a complete cognitive value chain, enabling more efficient utilization of textual information.
[0077] In the theoretical model and experimental scenario, it is assumed that existing methods output only one summary (approximately 0.1L) for a text of length L, while this invention outputs four layers of results for the same text, with a total output length of approximately 0.4L–0.6L, where: Information layer: approximately 0.15L; Knowledge layer: approximately 0.15L; Cognitive layer: approximately 0.1L; Wisdom layer: approximately 0.1L; With the same text length, the amount of information output by this invention can be increased to 4–6 times that of existing methods, thereby achieving higher information utilization efficiency with the same computing power cost.
[0078] The output of this invention can be in the form of structured JSON, which includes clear hierarchical numbers, concept nodes, relation edges and reasoning structures, which is beneficial for subsequent automatic retrieval, knowledge graph construction and machine learning training.
[0079] Existing technologies struggle to output differentiated content based on users' cognitive levels, professional backgrounds, and application scenarios. However, this invention, through step S5, performs contextualized reasoning on cognitive layer results based on user demand parameters and application scenario constraints to generate wisdom layer results, thereby achieving multi-perspective cognitive matching and personalized learning services.
[0080] For beginners: The system prioritizes outputting information and simplified knowledge layers to help them quickly understand basic facts. For advanced learners: The system outputs complete knowledge and cognitive layers to help them build knowledge structures and logical frameworks. For decision-makers or researchers: The system outputs wisdom-level results, providing suggestions, decision-making basis, and action strategies. The system can provide different levels of content to different users based on the same input text.
[0081] This invention can extract suitable learning content from cognitive results based on the user's cognitive level and learning goals, forming a personalized learning path, including learning stages, topics, recommended content, sequence, and expected goals, thereby improving learning efficiency and relevance.
[0082] In steps S1–S5, this invention can make full use of pre-trained language models and large language models to achieve text parsing, information extraction, knowledge summarization, cognitive framework construction, and contextualized reasoning, forming an scalable multi-layered cognitive generation framework.
[0083] This invention allows for the selection of language models of different sizes (lightweight, medium-sized, and large-scale); it can be driven by a combination of rule-based models, statistical models, and language models; and it can be adapted and fine-tuned for specific fields (such as medicine, law, and education). It can add four layers of output modules—information, knowledge, cognition, and wisdom—to existing NLP systems or large language model services; and it supports extended applications across multiple documents, languages, and scenarios.
[0084] Through multi-layered output and conflict resolution mechanisms, the system's tolerance for noisy text and complex semantics is improved, reducing errors and biases that may occur in a single summary or a single question and answer session.
[0085] This embodiment employs the following system architecture to implement the method of the present invention: The text acquisition module retrieves target text data from the user interface or file storage.
[0086] The text parsing module performs sentence segmentation, word segmentation, entity recognition, topic recognition, and semantic structure analysis on the text to generate a basic text representation.
[0087] The information extraction module extracts facts, events, attributes, relationships, and contextual items from the basic text representation to generate information layer results.
[0088] The knowledge construction module clusters, deduplicates, associates, and summarizes the information layer results to construct knowledge units and generate knowledge layer results.
[0089] The cognitive organization module performs hierarchical organization, conflict resolution, perspective alignment, and semantic compression of knowledge units to generate cognitive layer results.
[0090] The intelligent output module combines user requirement parameters and application scenario constraints to perform contextualized reasoning on the cognitive layer results and generate intelligent layer results.
[0091] The multi-layer output module integrates the results from the information layer, knowledge layer, cognition layer, and wisdom layer into a multi-level output and returns it to the user.
[0092] The modules mentioned above are connected via data streams: Text Acquisition Module → Text Parsing Module → Information Extraction Module → Knowledge Construction Module → Cognitive Organization Module → Wisdom Output Module → Multi-layer Output Module.
[0093] The data transmitted between modules includes: basic text representation, information layer results, knowledge layer results, cognition layer results, wisdom layer results, and multi-level output structures.
[0094] The input text in this example is a short article about "health education policies in a certain city," the content of which is as follows (simplified example): "In 2024, a certain city issued the 'Health Education Promotion Policy,' stipulating that starting in 2025, all primary and secondary schools must arrange at least two health education classes per week, covering topics such as diet and nutrition, mental health, and exercise. The policy requires schools to cooperate with community health service centers to conduct health lectures and physical examinations. The policy aims to improve the health literacy of adolescents and reduce obesity and chronic disease rates. The policy is expected to invest 50 million yuan in curriculum development, teacher training, and community cooperation." Step S1: Text Acquisition and Basic Representation Construction (Static Relations and Dynamic Relations) The text parsing module contains components such as a sentence segmenter, word segmenter, entity recognizer, topic recognizer, and semantic parser.
[0095] Static Relationship: The input text is a string containing several sentences and paragraphs.
[0096] Dynamic relationships: 1. The text acquisition module reads the above text from the user interface and performs encoding standardization and cleaning.
[0097] 2. The text parsing module segments the text into sentences, resulting in approximately 6 sentences.
[0098] 3. Perform word segmentation and part-of-speech tagging on each sentence to obtain a word sequence.
[0099] 4. Perform entity recognition to identify entities such as "a certain city", "2024", "Health Education Promotion Policy", "2025", "primary and secondary schools", and "community health service centers".
[0100] 5. Conduct thematic identification and identify keywords such as "health education", "policy", "school", "community" and "health literacy".
[0101] 6. Perform semantic structure analysis to identify subject-verb-object structures and causal relationships (such as "the policy objective is to increase... and decrease...").
[0102] 7. Use a pre-trained language model to generate sentence vectors and full-text vectors.
[0103] This process transforms raw text into a structured textual representation, encompassing sentences, words, entities, topics, semantic structures, and vector representations. It provides a unified and actionable input for subsequent extraction of information, knowledge, cognition, and intelligence.
[0104] Step S2: Information layer result generation (static and dynamic relationships) Static relationships: The information layer results include lists of fact items, event items, attribute items, relationship items, and context items. Each information item contains attributes such as type, content, source sentence, time, and location.
[0105] Dynamic relationships: 1. The information extraction module extracts factual items from the basic text representation, such as "A certain city released the 'Health Education Promotion Policy' in 2024" and "Starting from 2025, all primary and secondary schools will arrange at least 2 health education courses per week"; 2. Extract event items, for example, event type: policy release; participants: a city, primary and secondary schools, community health service centers; time: from 2024 and 2025; process: policy release, course arrangement, lectures and physical examinations; results: improved health literacy, reduced obesity rate and incidence of chronic diseases; 3. Extract attribute items, such as "at least 2 health education classes per week" and "estimated investment of 50 million yuan"; 4. Extract related terms, such as "cooperation between schools and community health service centers" and "the policy goal is to improve the health literacy of adolescents"; 5. Extract contextual items, for example: policy background: improve health literacy among teenagers; constraints: at least 2 classes per week; resource constraints: investment of 50 million yuan; This process generates information layer results, answering the question "What does the text say?" and representing the raw facts and explicit information within the text. It provides foundational factual input for subsequent knowledge, cognition, and wisdom layer refinement.
[0106] Step S3: Knowledge Layer Result Generation (Static Relationships and Dynamic Relationships) Static relationships: The results of the knowledge layer include concept associations, rule summaries, method abstractions, and experience inductions. Each knowledge unit contains: concept nodes, relationship edges, rule descriptions, method sequences, and experience descriptions, etc.
[0107] Dynamic relationships: 1. The knowledge construction module clusters information items, grouping items with similar semantics or the same theme into one category to form several information clusters, such as "policy content cluster", "implementation measures cluster", "policy objectives cluster" and "resource input cluster".
[0108] 2. Remove duplicate information within the same cluster and merge semantically similar information items.
[0109] 3. Extract core concepts from information clusters to form concept nodes, such as "health education policy", "primary and secondary schools", "health education curriculum", "health literacy", "obesity rate", "chronic diseases", etc.
[0110] 4. Extract the relationships between concepts from relational terms and event terms to form relational edges, for example, "Health education policy → Requirement → Primary and secondary schools to offer health education courses", "Health education courses → Help improve → Health literacy", "Improved health literacy → Helps reduce → Obesity rate and incidence of chronic diseases"; 5. Summarize multiple information clusters and identify patterns, such as "systematic health education policies + school-community cooperation → help improve health literacy" and "at least 2 health education classes per week → help reduce obesity rates"; 6. Abstract methods from event items and attribute items. For example, method name: Health Education Course Implementation Method; Step sequence: 1) Develop policies; 2) Arrange courses; 3) Conduct lectures and physical examinations; 4) Evaluate the effects; 7. Summarize experiences from information sets, for example, "Under conditions of sufficient resources (such as an investment of 50 million yuan), health education policies are more likely to achieve their expected goals"; This process generates knowledge-level results, answering the question "What pattern does the text illustrate?" and achieving the extraction of knowledge from information. It also constructs knowledge units and concept-relationship networks, providing a foundation for the subsequent organization of the cognitive-level framework.
[0111] Step S4: Generate cognitive layer results (static and dynamic relationships) Static relationships: Cognitive layer outcomes include understanding structures, judgment structures, and reasoning structures. Cognitive frameworks include hierarchical frameworks, causal frameworks, contrastive frameworks, temporal frameworks, and question-answer frameworks.
[0112] Dynamic relationships: 1. The cognitive organization module selects the cognitive framework type based on the theme and structure of the knowledge layer results: Hierarchical framework: Represents the hierarchical relationship of "policy → curriculum → health literacy"; Causal Frame: Represents the causal relationship of "health education policy → improved health literacy → reduced obesity rate and chronic disease"; Timeline framework: This indicates the chronological order of "policy release in 2024 → curriculum implementation starting in 2025 → long-term reduction in obesity rate"; Question-answer framework: "How to improve health literacy among adolescents? → Through the implementation of health education policies and curricula."
[0113] 2. Organize knowledge units according to hierarchical relationships to form a hierarchical structure, such as: Top-level: Health education policy; Middle level: Curriculum implementation, school-community collaboration; Basic level: health literacy, obesity rate, chronic diseases; 3. Extract causal relationships from event items and pattern summaries to construct causal chains, such as: Reasons: Issuing health education policies + arranging courses + collaborating on lectures and physical examinations; Result: Improved health literacy led to a decrease in obesity rates and the incidence of chronic diseases; 4. Arrange events in chronological order to form a timeline: 2024: Policy release; Starting in 2025: Curriculum will be implemented; Long-term: Reduces obesity rates and chronic diseases; 5. Identify questions and answers to form question-answer pairs: Question: How can we improve the health literacy of teenagers? Answer: Through the implementation of health education policies and curricula; 6. Resolve conflicts in contradictory information within knowledge units (there are no obvious contradictions in this example), calculate confidence levels, and form a knowledge set with good consistency.
[0114] 7. Perform semantic compression on knowledge units to form high-level summaries and understanding frameworks, such as: "A certain city has issued a health education policy, requiring primary and secondary schools to offer health education courses and to cooperate with communities to conduct lectures and physical examinations, with the aim of improving health literacy and reducing obesity and chronic disease rates." This process generates cognitive-level results, answering the question "What understanding framework does the text form?" and achieving the distillation from knowledge to cognition. Constructing a cognitive framework diagram facilitates the visualization and understanding of the text's overall logic.
[0115] Step S5: Generation of Wisdom Layer Results (Static Relationships and Dynamic Relationships) Static relationships: The results of the intelligence layer include suggestions, decision-making references, action strategies, and personalized learning paths. User requirement parameters include: user cognitive level, learning objectives, professional background, application scenario, and output granularity. Application scenario constraints include: time constraints, resource constraints, risk constraints, and policy constraints.
[0116] Dynamic relationships: 1. The intelligent output module obtains user requirement parameters from user input or system settings, such as user cognitive level: beginner; learning goal: understand basic concepts; professional background: education; application scenario: learning; output granularity: summary; 2. Obtain application scenario constraints, such as time constraints: learning must be completed within one week; resource constraints: available school and community resources; policy constraints: must comply with local education policies; 3. Based on the results of the cognitive layer, combined with user needs parameters and application scenario constraints, contextualized reasoning is performed: the cognitive framework is mapped to a specific context (education field); the output depth is adjusted according to the user's cognitive level (beginner), prioritizing the provision of the information layer and the simplified knowledge layer; relevant knowledge and methods are selected based on professional background (education); and targeted suggestions are generated based on the application scenario (learning).
[0117] 4. Generate suggestions, such as: "It is recommended that educators pay attention to this policy and understand the curriculum content and implementation requirements," and "It is recommended that schools plan their curriculum and teacher training in advance." 5. Generate decision-making references, such as: "Decision basis: Policy requires at least 2 classes per week, resources need to be planned in advance", "Balance factors: Class time vs. other subject time", "Risk points: Insufficient teachers, inconsistent course quality"; 6. Generate action strategies, for example, strategy name: Health Education Curriculum Implementation Strategy; Step sequence: 1) Develop a curriculum plan; 2) Train instructors; 3) Collaborate with community health service centers; 4) Conduct lectures and physical examinations; 5) Evaluate effectiveness and make adjustments; 7. Generate personalized learning paths, for example: Learning Phase 1: Understand the policy background and objectives; Learning Phase 2: Learn the course content and implementation methods; Learning Phase 3: Understand the school-community cooperation model; Learning Phase 4: Evaluate the policy effect; Recommended content: original policy text, course guide, case sharing.
[0118] The goal is to generate a layer of wisdom that answers the question, "How should we choose or act when faced with this type of problem?" This process elevates cognition to wisdom. It provides targeted suggestions, decision-making references, action strategies, and personalized learning paths for different users, guiding their choices and actions in specific situations.
[0119] Step S6: Constructing multi-level outputs (static and dynamic relationships) Static relationships: Multi-level output includes hierarchical structure, headings, numbering, and content sets. Output formats can be selected as tree structure, hierarchical list, structured JSON, or combined text.
[0120] Dynamic relationships: 1. The multi-layer output module organizes the results of the information layer, knowledge layer, cognition layer and wisdom layer into a hierarchical structure according to the order of cognitive hierarchy.
[0121] 2. Select Structured JSON as the output format, which includes: level: information / knowledge / cognition / wisdom • items: List of results for each level • hierarchy: hierarchical numbering and relationships 3. Add a title, number, and brief description to each level to ensure that the hierarchical relationship is clear.
[0122] 4. Conduct a consistency check to ensure that there are no duplicates or conflicts in the content at each level and that the hierarchical relationships are clear.
[0123] 5. Return multi-level output to the user interface and record the output log.
[0124] This allows for the extraction and output of multiple cognitive results from the same input text, meeting the needs of multi-perspective cognitive matching and personalized learning services. It also improves the efficiency of text information utilization and the interpretability of results.
[0125] The modules used in this embodiment, such as text parsing, information extraction, knowledge construction, cognitive organization, and intelligent output, can all be implemented based on existing NLP tools, pre-trained language models, and large language models.
[0126] Mature algorithms and tools already exist for steps such as sentence segmentation, word segmentation, entity recognition, topic recognition, and semantic parsing.
[0127] The steps of information clustering, concept node construction, relation edge construction, pattern summarization, method abstraction, and experience induction can be achieved by combining statistical methods and language models.
[0128] Steps such as cognitive framework construction, conflict resolution, perspective alignment, and semantic compression can be achieved by combining rule models and language models.
[0129] The steps of contextualized reasoning, suggestion generation, decision reference generation, action strategy generation, and personalized learning path generation can be jointly implemented based on a large language model and a rule model.
[0130] This embodiment can adopt a modular design, and each module can be upgraded and replaced independently. It can support extended applications with multiple documents, multiple languages, and multiple scenarios. It can be adapted and fine-tuned for specific fields (such as medicine, law, and education).
[0131] The input text can be any article, report, news, paper, document, etc., with a wide range of data sources. Pre-trained language models and large language models are publicly available, requiring no additional large-scale data training.
[0132] This embodiment can obtain four layers of results in a single process, significantly reducing computational resource consumption and labor costs compared to existing technologies that require multiple independent processes. In the theoretical model, processing costs can be reduced by approximately 60%–70%.
[0133] This embodiment outputs structured JSON, containing explicit hierarchical numbers, concept nodes, relational edges, and reasoning structures, facilitating subsequent automatic retrieval, knowledge graph construction, and machine learning training. In theoretical experiments, user comprehensibility scores improved from 2.5–3.0 points in existing methods to 3.8–4.5 points.
[0134] This embodiment can generate differentiated outputs and personalized learning paths based on the user's cognitive level, professional background, and application scenario. In theoretical experiments, user satisfaction scores increased from 2.8–3.2 points in existing methods to 3.8–4.5 points, and learning efficiency improved by approximately 30%–50%.
[0135] This embodiment can generate different levels of content for different learners, supporting the construction of personalized education and lifelong learning systems. In scenarios such as policy analysis, health management, and public education, it can generate decision-making references and action strategies to support scientific decision-making and refined governance.
[0136] To verify the technical effect of the present invention, the following effect experiment was designed in this embodiment, including the experimental process, results and data analysis.
[0137] Experimental subjects: 20 participants, including 10 beginners (non-professional background) and 10 advanced learners (education or related professional background).
[0138] Experimental data: 10 texts on different topics, including health education policies, environmental policies, science and technology news, and academic paper abstracts.
[0139] The design involved two groups: a control group, which used existing single-summary methods to output a single summary for each text; and an experimental group, which used the method of this invention to output four layers of results for each text: information layer, knowledge layer, cognitive layer, and wisdom layer.
[0140] Evaluation Indicators: 1. Information Content Score (1–5 points): Evaluates the richness of the output information. 2. Comprehensibility Score (1–5 points): Evaluates the ease of understanding the output. 3. User Satisfaction Score (1–5 points): Evaluates overall user satisfaction. 4. Learning Efficiency: Records the time participants take to complete related test questions after reading the output and calculates the average learning time. 5. Test Question Accuracy: Records the accuracy rate participants achieve when completing related test questions.
[0141] Twenty participants were randomly divided into 10 groups of two (one beginner and one advanced learner) each. Each group was randomly assigned 10 texts, with one text processed by the control group and the other by the experimental group. After reading the output, participants completed five related test questions and recorded their completion time. Participants rated the output for information content, comprehensibility, and user satisfaction. Rating and test data were collected and statistically analyzed.
[0142] The experimental results (average values) are shown in Table 1: ; Table 1 The information content score of the experimental group (4.2) was significantly higher than that of the control group (2.9), indicating that the present invention outputs richer information and can provide more dimensions of information.
[0143] The comprehensibility score of the experimental group (4.0) was significantly higher than that of the control group (2.7), indicating that the four-layer output structure of the present invention is clear and helps users understand the overall logic of the text.
[0144] The user satisfaction score of the experimental group (4.3) was significantly higher than that of the control group (3.0), indicating that the present invention can better meet the user's needs for multi-perspective cognition and personalized learning.
[0145] The average learning time of the experimental group (10 minutes) was shortened by about 33% compared with that of the control group (15 minutes), indicating that the personalized learning path and hierarchical output of the present invention help improve learning efficiency.
[0146] The test accuracy rate of the experimental group (82%) was significantly higher than that of the control group (68%), indicating that the output of this invention helps users better understand and apply the text content.
[0147] Experimental results show that the present invention significantly outperforms existing single-summary methods in terms of information content, comprehensibility, user satisfaction, learning efficiency, and test question accuracy, verifying the technical effectiveness of the present invention in the following aspects: It enables multiple cognitive outputs corresponding to the same input text; it constructs a cognitive hierarchy to improve the interpretability and reusability of results; and it supports multi-perspective cognitive matching and personalized learning services to improve user satisfaction and learning efficiency.
Claims
1. A method for extracting information value based on cognitive hierarchy progression, characterized in that, Includes the following steps: S1. Obtain the target text data to be processed, and perform sentence segmentation, word segmentation, entity recognition, topic recognition and semantic structure parsing on the target text data to obtain the basic text representation; S2. Based on the text basic representation, extract fact items, event items, attribute items, relationship items and context items in the text to generate information layer results for representing the original facts and explicit information in the text; S3. Based on the information layer results, perform knowledge aggregation and rule induction, cluster, deduplicate, associate and merge multiple information elements, construct knowledge units, and generate knowledge layer results to represent the rules, methods and experiences extracted from the information. S4. Based on the knowledge layer results, construct a cognitive framework, perform hierarchical organization, conflict resolution, perspective alignment and semantic compression of knowledge units, and generate cognitive layer results to represent the understanding structure, judgment structure or reasoning structure of the target text. S5. Based on the cognitive layer results, combined with application scenario constraints and user demand parameters, contextual reasoning is performed on the cognitive layer results to generate intelligent layer results for outputting suggestions, decision references, action strategies, or personalized learning paths. S6. Return the information layer results, knowledge layer results, cognition layer results and wisdom layer results as multi-level outputs to achieve the extraction and output of multiple cognitive results corresponding to the same input text.
2. The method according to claim 1, characterized in that, The text-based representation includes one or more of the following: text vector representation, paragraph structure representation, keyword representation, and discourse logic representation. The knowledge layer results include one or more of the following: concept association, rule summary, method abstraction, and experience induction. The cognitive framework includes one or more of the following: hierarchical framework, causal framework, contrastive framework, temporal framework, and question-answer framework; The user requirement parameters include one or more of the following: user cognitive level, learning objectives, professional background, application scenario, and output granularity. The multi-level output can be one or more of the following: tree structure, hierarchical list, structured JSON, or combined text.
3. The method according to claim 1, characterized in that, Step S1 specifically includes: Obtain the target text data. If the text is a collection of multiple documents, process them sequentially according to document number or time order, or merge multiple documents and process them uniformly. At the same time, unify the encoding format of the text and remove garbled characters and invisible characters. Remove noise from the text, including HTML tags, advertisements, headers and footers, and irrelevant symbols; compress redundant whitespace characters, compressing multiple consecutive spaces into a single space and multiple consecutive line breaks into a single line break, while standardizing punctuation in Chinese text; The text is segmented into sentences based on sentence end markers, long sentences are broken down according to semantic connectors, and the start and end positions of each sentence are recorded to form a sentence list; Each sentence is segmented into words to obtain a word sequence, and each word is tagged with its part of speech. Named entity recognition is performed on the word sequence, each entity is labeled with its type, and its position in the text is recorded; entities are bound to corresponding sentences to form an entity-sentence association structure; the entities include personal names, place names, organization names, time, numbers, and technical terms; Based on keyword statistics, topic models, or pre-trained language models, identify the keywords and core topics of the text, extract the keywords of each paragraph, summarize them into a set of keywords for the whole text, and mark the topic category to which each sentence belongs; Perform dependency parsing or semantic role labeling on sentences to identify subject-verb-object structures, modification relations, causal relations and conditional relations, construct semantic relations between sentences, and record logical relations between paragraphs; Each sentence and the entire text are input into a pre-trained language model to obtain sentence vector representations and full-text vector representations. These vector representations are then associated with sentences, paragraphs, entities, and topics to form a complete basic text representation.
4. The method according to claim 1, characterized in that, Step S2 specifically includes: The process extracts fact items by identifying declarative facts from the text's basic representation, recording the fact type, subject, time, location, and state, and binding them to corresponding sentences to form fact-sentence relationships. It also extracts event items by identifying events from the text's basic representation, recording the event type, participants, time, location, process, and result, and binding them to corresponding sentences and paragraphs. Furthermore, it extracts attribute items by extracting object attributes from the text's basic representation, recording the extracted object, attribute name, attribute value, and data source sentence, and binding them to corresponding entities and sentences. Finally, it extracts relationship items by identifying relationships between entities, recording the subject, relationship type, object, and source sentence, and binding them to corresponding entities and sentences. Finally, it extracts contextual items by identifying contextual information in the text, recording the context type, content, scope of application, and source sentence, and binding them to corresponding paragraph entities. The fact items, event items, attribute items, relationship items, and context items are integrated into a unified information layer result structure, and duplicate or conflicting information is initially marked for subsequent processing steps.
5. The method according to claim 1, characterized in that, Step S3 specifically includes: Clustering is performed on fact items, event items, attribute items, relationship items, and context items. Items with similar semantics or the same theme are grouped together to form several information clusters. Each cluster corresponds to a potential knowledge theme. Clustering is based on topic similarity, entity overlap, and vector distance. Duplicate information within the same cluster is removed, conflicting information is marked, and information items with the same or highly similar semantics are merged to form a simplified information set. Core concepts are extracted from the information set to form concept nodes. For each concept node, the concept name, definition, related attributes, and source information items are recorded, and the concept node is associated with the corresponding information item. Extract the relationships between concepts from relation items and event items to form relation edges. For each relation edge, record the source concept, relation type, target concept, and source information item to construct a concept relation network. Summarize multiple information clusters, identify recurring patterns, trends or regularities, and summarize and record the regularity description, applicable conditions, source information cluster and confidence level for each regularity to form a regularity summary list; Reusable methods are abstracted from event items and attribute items. For each method, the method name, step sequence, applicable scenario and source information are recorded to form a list of abstract methods. Summarize empirical conclusions from the information set, and record the empirical description, applicable conditions, source information and confidence level for each empirical summary to form an empirical summary list; The conceptual relationship network, the list of rules, the list of abstract methods, and the list of inductive experience are integrated into a unified knowledge layer result.
6. The method according to claim 1, characterized in that, Step S4 specifically includes: Based on the theme and structure of the knowledge layer results, select the cognitive framework type, including hierarchical framework, causal framework, contrastive framework, temporal framework and question-answer framework; Within the hierarchical framework, knowledge units are organized according to hierarchical relationships to form a hierarchical structure. Each level records its level number, concept set, and relationship type, forming a knowledge hierarchy tree. Within the causal framework, causal relationships are extracted from event items and rule summaries to construct causal chains. For each causal chain, the cause event, result event, causal relationship type, and confidence level are recorded, forming a causal graph structure. Within the comparative framework, different objects, solutions, or viewpoints are compared to form a comparative structure. For each comparative item, the comparison object, comparison dimension, differences, and sources are recorded, forming a comparison matrix. Within the temporal framework, events are arranged chronologically to form a temporal structure. For each point in time, time, events, and state changes are recorded, forming a timeline. Within the question-answer framework, questions and their corresponding answers are identified from text, forming question-answer pairs. For each question-answer pair, the question description, answer content, and source sentence are recorded, forming a question-answer list. Conflict resolution is performed on contradictory information in knowledge units, the confidence level of each piece of information is calculated, and priority is determined based on source authority, time sequence, and semantic consistency. Low-priority information is corrected or deleted, or marked as "disputed", thus forming a consistent knowledge set. Align different perspectives on the same topic, identify different perspectives, group knowledge units according to perspective, and form a multi-perspective knowledge structure. Semantic compression of knowledge units is performed to form high-level summaries and understanding frameworks. Core concepts and key relationships are extracted, compressed into simplified cognitive structures, and understanding structure descriptions are formed. Understanding structures are numbered and labeled to form cognitive framework diagrams. This integrates the understanding structure, judgment structure, and reasoning structure into a unified cognitive layer result.
7. The method according to claim 1, characterized in that, Step S5 specifically includes: Obtain user demand parameters and application scenario constraints, including time constraints, resource constraints, risk constraints, and policy constraints; Based on the cognitive layer results, combined with user demand parameters and application scenario constraints, contextualized reasoning is performed to map the cognitive framework to specific situations. The output depth is adjusted according to the user's cognitive level, relevant knowledge and methods are selected based on professional background, and targeted suggestions or strategies are generated based on the application scenario to form contextualized reasoning results. Extract suggestions from contextualized reasoning results, and record the content, applicable conditions, source of evidence, and confidence level of each suggestion to form a suggestion list; Extract decision references from contextualized reasoning results, and record the decision problem, basis, trade-off factors, risk points and sources for each decision reference to form a decision reference list; Extract action strategies from contextualized reasoning results, including step sequences, key points of operation, and precautions. Record the strategy name, step sequence, applicable scenario, and precautions for each action strategy to form an action strategy list. Based on the user's cognitive level and learning goals, suitable learning content is extracted from the cognitive layer results to form a personalized learning path. The learning path records the learning stages, learning topics, recommended content, learning order and expected goals. The suggestion list, decision reference list, action strategy list, and personalized learning path are integrated into a unified intelligent layer result.
8. The method according to claim 1, characterized in that, Step S6 specifically includes: The results of the information layer, knowledge layer, cognition layer and wisdom layer are organized into a hierarchical structure according to the order of cognitive hierarchy. The hierarchical name, hierarchical number and content set of each level are recorded to form a cognitive hierarchy tree. Based on user needs and application scenarios, select the output format, including tree structure, hierarchical list, structured JSON, and combined text, and parameterize the output format; Integrate the results from each level into a multi-level output in the selected format, and add a title, number, and brief description to each level to ensure a clear hierarchical relationship between the levels; Perform consistency checks on multi-level outputs.
9. An information value extraction system based on the method of any one of claims 1-8, characterized in that, include: The text acquisition module retrieves target text data from the user interface or file storage; The text parsing module performs sentence segmentation, word segmentation, entity recognition, topic recognition, and semantic structure analysis on the text to generate basic text representations. The information extraction module extracts facts, events, attributes, relationships, and contextual items from the basic text representation to generate information layer results; The knowledge construction module clusters, deduplicates, associates, and summarizes the information layer results to build knowledge units and generate knowledge layer results. The cognitive organization module performs hierarchical organization, conflict resolution, perspective alignment, and semantic compression of knowledge units to generate cognitive layer results. The intelligent output module combines user demand parameters and application scenario constraints to perform contextualized reasoning on the cognitive layer results and generate intelligent layer results. The multi-layer output module integrates the results from the information layer, knowledge layer, cognition layer, and wisdom layer into a multi-level output and returns it to the user.
10. The system according to claim 9, characterized in that: The application scenario constraints include time constraints, resource constraints, risk constraints, and policy constraints; the user requirement parameters include user cognitive level, learning objectives, professional background, application scenario, and output granularity.
