CN-121998044-A - Cognitive process visualization method, system, equipment and medium oriented to human-computer interaction

CN121998044ACN 121998044 ACN121998044 ACN 121998044ACN-121998044-A

Abstract

The invention provides a cognitive process visualization method, a system, equipment and a medium for human intelligence interaction, wherein the method comprises the steps of acquiring a natural language interaction dialogue with an artificial intelligence AI, and carrying out high-dimensional semantic analysis and cognitive relation analysis to convert the natural language interaction dialogue into a structured reasoning event object; the method comprises the steps of determining a structured cognitive chain for revealing the action of a thinking path and an AI according to an reasoning event object through multi-dimensional relation analysis and a stage self-adaptive rule, determining a cognitive input index and an creative index according to the structured cognitive chain and quantizing node semantics, structures and interaction modes, and generating a data-driven multi-level visual interface according to the cognitive input index and the creative index by combining the structured cognitive chain to perform automatic layout and visual coding. The invention realizes data-driven multi-level visualization, so that teachers can realize visualization, diagnosis and intervention of the learning process in an AI-supported open learning task.

Inventors

Zha Siyu
LIU YUJIA
GONG JIANGTAO
XU YINGQING

Assignees

清华大学

Dates

Publication Date: 20260508
Application Date: 20251224

Claims (10)

1. A cognitive process visualization method oriented to human-intelligence interaction is characterized by comprising the following steps: Acquiring a natural language interactive dialogue with an artificial intelligence AI, and performing high-dimensional semantic analysis and cognitive relation analysis to convert the natural language interactive dialogue into a structured reasoning event object; according to the reasoning event object, determining a structured cognitive chain for revealing the action of a thinking path and an AI through multi-dimensional relation analysis and stage self-adaptive rules; According to the structured cognitive chain, node semantics, structure and interaction modes are quantized, and cognitive input indexes and creative indexes are determined; and according to the cognitive input index and the creative index, combining the structured cognitive chain to perform automatic layout and visual coding, and generating a data-driven multi-level visual interface.
2. The human-intelligence interaction oriented cognitive process visualization method of claim 1, wherein obtaining a natural language interaction dialogue with an artificial intelligence AI and performing high-dimensional semantic parsing and cognitive relationship analysis to convert the natural language interaction dialogue into a structured inference event object, comprises: acquiring a natural language interactive dialogue with an artificial intelligence AI, and performing syntactic structure decomposition based on an attention mechanism to obtain semantic question fragment characteristics corresponding to student expression and semantic reply content characteristics corresponding to AI reply; classifying the semantic question segment features corresponding to the student expressions, and performing function type recognition on the semantic reply content features corresponding to the AI replies to obtain corresponding semantic segment categories and corresponding function types; According to the semantic question fragment characteristics corresponding to the student expression and the semantic reply content characteristics corresponding to the AI reply, similarity matching, reference pattern recognition and logic dependency analysis are carried out, and an AI influence factor is determined, wherein the AI influence factor is used for representing the influence degree of the AI reply on the student; According to the semantic question segment features, the semantic segment categories, the semantic reply content features and the function types corresponding to the AI replies, task context prediction is carried out to obtain a task prediction stage; and obtaining a corresponding reasoning event object according to the semantic question fragment characteristics corresponding to the student expression, the semantic fragment category, the semantic reply content characteristics corresponding to the AI reply, the function type, role information, the timestamp and the task prediction stage, determining the context attachment relationship with the previous other reasoning event objects, and updating the reasoning event object.
3. The method for visualizing a human-intelligence interaction oriented cognitive process of claim 1, wherein said inference event object comprises a semantic question segment feature, a semantic segment class, a semantic reply content feature, a function type, character information, a time stamp, and a task prediction stage corresponding to an AI reply expressed by a student, wherein said determining a structured cognitive chain for revealing a mental path and an AI effect according to said inference event object by multidimensional relation analysis and stage adaptation rules comprises: Determining semantic similarity with other prior reasoning event objects according to the reasoning event objects; Carrying out causal relationship identification according to the reasoning event object and other previous reasoning event objects; according to the semantic segment category and task context prediction of the reasoning event object, carrying out theme consistency judgment; Carrying out inference action continuity assessment according to semantic segment categories and AI influence factors in the inference event objects; And selecting a corresponding reasoning rule according to a task prediction stage corresponding to the reasoning event object, and determining a structural relation among the reasoning event objects by combining semantic similarity, causal relation, main consistency and reasoning action consistency to form a structured cognitive chain with directivity and a hierarchical structure, wherein each node in the structured cognitive chain represents a logic state of a student or an AI in a corresponding round of interaction, and the connection between the nodes is used for representing a reasoning migration path from divergence to convergence, scheme exploration to scheme screening and interpretation to disbelief.
4. A cognitive process visualization method for human-intelligence interactions as claimed in claim 3, comprising, after determining a structured cognitive chain for revealing mental paths and AI actions from the inference event objects by multi-dimensional relational analysis and phase adaptation rules: according to the topological structure and semantic features of the structured cognitive chain, analyzing text density, interpretation depth, node expansion speed, content repeatability, semantic vector similarity and chain topological change, identifying learning risks, marking risk nodes and recording risk types.
5. A cognitive process visualization method for human-intelligence interactions as claimed in claim 3, comprising, after determining a structured cognitive chain for revealing mental paths and AI actions from the inference event objects by multi-dimensional relational analysis and phase adaptation rules: According to the structured cognitive chain, semantic behavior characteristics and reasoning action analysis are carried out, students are determined to be in the divergent, convergent, planning, executing or thinking-back stages, and stage labels are superimposed into the structured cognitive chain, so that stage judgment is correlated with reasoning structure and index calculation.
6. The human-intelligence interaction oriented cognitive process visualization method of claim 1, wherein determining cognitive input and creative indicators from the structured cognitive chain, quantifying node semantics, structure and interaction patterns, comprises: According to the semantic question fragment features and the semantic reply content features in the structured cognitive chain, carrying out semantic deep analysis on each node, and determining the semantic complexity of the content of the corresponding node; Performing inference action type evaluation according to semantic segment categories in the structured cognitive chain to obtain action inference types; According to the connection relation between the student nodes and the AI nodes in the structured cognitive chain and the AI influence factors, performing interactive structure judgment, and determining a cognitive input index by combining the semantic complexity of the node content and the action reasoning type; according to the structured cognitive chain, determining the number of nodes of the student for generating ideas through semantic clustering, semantic span analysis and chain structural feature recognition; Carrying out semantic clustering on all the idea nodes to obtain a clustering result and determining category diversity; according to the structured cognitive chain, semantic span analysis is carried out, semantic distances between each idea node of the student and the corresponding views of the AI answer are determined, and originality is obtained; Determining the divergence degree of the chain based on the chain branch structure according to the structured cognitive chain, and determining the extension elaboration depth through the logic chain length and the reason depth interpreted by the nodes; and determining creative indexes according to the node number of the student generated ideas, the category diversity, the originality and the elaboration depth.
7. The method for visualizing a human-intelligence interaction oriented cognitive process of claim 1, wherein said multi-level visual interface comprises a visual overview presentation and a student character level presentation, wherein said automatically laying out and visually encoding in combination with said structured cognitive chain according to said cognitive input index and creative index, generates a data-driven multi-level visual interface comprising: according to the structured cognitive chain, the cognitive input index and the creative index, key indexes corresponding to student roles are integrated, and students with learning risks are identified; Generating visual overview display according to the key indexes of the integrated student roles and the students with learning risks, wherein the visual display displays the student roles in the whole class or a preset range in a list or grid form, and visually presents the task stage, input trend and risk state of each person by using color codes, small icons or mini charts; generating an event sequence curve according to the cognitive input index and the creative index and the time sequence of a dialogue for any student role, or superposing and presenting the cognitive input index and the creative index on the structured cognitive chain; And determining the node position according to the topological structure of the structured cognitive chain, enabling the main path to be positioned in the center of the view, expanding the branch paths in a hierarchical manner, performing visual coding to determine the node color, transparency, line type and mark and highlight the risk nodes according to the node attribute, and generating student role level display.
8. A cognitive process visualization device oriented to human-computer interaction, comprising: The object conversion module is used for acquiring a natural language interaction dialogue with the artificial intelligence AI and carrying out high-dimensional semantic analysis and cognitive relation analysis so as to convert the natural language interaction dialogue into a structured reasoning event object; The knowledge chain generation module is used for determining a structured cognitive chain for revealing the action of the thinking path and the AI through multi-dimensional relation analysis and stage self-adaptive rules according to the reasoning event object; The index quantization module is used for determining a cognitive input index and a creative index according to the structural cognitive chain, the node semantics, the structure and the interaction mode; And the visual processing module is used for carrying out automatic layout and visual coding by combining the structured cognitive chain according to the cognitive input index and the creative index to generate a data-driven multi-level visual interface.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and running on the processor, characterized in that the processor implements the cognitive process visualization method oriented towards human-intelligent interactions as claimed in any one of claims 1 to 7 when executing the computer program.
10. A non-transitory computer readable storage medium having stored thereon a computer program, wherein the computer program when executed by a processor implements the human-intelligence interaction oriented cognitive process visualization method of any of claims 1 to 7.

Description

Cognitive process visualization method, system, equipment and medium oriented to human-computer interaction Technical Field The invention relates to the technical field of artificial intelligence, in particular to a cognitive process visualization method, a system, equipment and a medium for human-intelligence interaction. Background With the breakthrough progress of the generated artificial intelligence technology, the application of the technology in the education field is deepened into cognitive partners capable of participating in a complex learning process from early intelligent question-answering and knowledge retrieval. In scenes such as project learning (PBL) and open tasks, the dialogue between students and AI has become an important medium for promoting tasks, understanding information, generating creative and solving problems, and the interaction mode breaks through the limitations of the traditional learning tools and provides personalized and instant support for the students through natural language interaction. However, deep application of the technology also brings new challenges, such dialogues are strong in unstructured nature, semantic jumping nature and high in degree of freedom, and key cognitive behaviors such as reasoning process, viewpoint transformation, decision logic and the like generated by students in interaction are scattered in the context of long sequences in natural language form, so that deep analysis and effective intervention on the learning process become the difficult problems to be solved urgently in the prior art. At present, the management of the interaction between students and AI is mainly realized through linear text record and simple task progress tracking of an Intelligent Tutoring System (ITS) or a Learning Management System (LMS), linear dialogue text and related metadata between the students and AI are completely recorded through a back-end database, a log is primarily processed by adopting a technology based on rules or shallow Natural Language Processing (NLP), for example, whether the students mention specific concepts or not is identified through keyword matching, dialogue emotion tendencies are judged through emotion analysis tools, or behavioral indexes such as the number of questioning, dialogue rounds, task completion states and the like are counted, so that analysis results are summarized into simple learning reports or data dashboards for teachers to view afterwards, and the core aim is to track the learning progress and participation degree of the students instead of analyzing the intrinsic cognitive processes of the students. However, the linear text recording mode completely loses the cognitive structure relation in the dialogue, can not present deduction, comparison, integration and correction among different ideas of students, and is difficult to reflect the activity curves of the students at each stage of the task, so that teachers can not accurately judge the real thinking process of the students in the task by checking dialogue contents in time, can not reveal the influence positions of AI answers in a student decision chain, and particularly can not identify whether the students replace AI suggestion machinery with own thinking or show meaningful cognitive processing, so that learning activities under AI support have obvious limitations in aspects of teaching controllability, learning quality assurance, process evaluation and the like, and if teachers try to manually analyze the contents, a great amount of time cost is required, key nodes are easy to ignore due to information redundancy, and the real thinking path of the students is difficult to grasp. In addition, the analysis method based on keywords and simple statistics is too superficial to quantify high-dimensional cognitive indexes such as cognitive input depth (constructivity, interactivity), creativity (fluency, originality) or integrity of inference chains, and evaluation results are subjective and lagged. In addition, as different students push tasks in different stages, different conversation lengths and different paths, the technology can only provide scattered individual data or coarse and shallow macroscopic progress, can not realize multi-level deep cognitive behavior visualization, and is difficult to dynamically position learning risk nodes, so that teachers are not careful in managing large-scale asynchronous AI learning activities, and teaching controllability and learning quality guarantee are severely limited. Disclosure of Invention The invention provides a cognitive process visualization method, a system, equipment and a medium oriented to human-intelligence interaction, which are used for solving the defect that a linear text recording mode in the prior art cannot reveal a true thinking path of students in tasks, realizing data-driven multi-level visualization, enabling teachers to realize visualization, diagnosis and intervention of a learning process in an AI-supp