Multi-layer cognitive corpus construction method, device, medium and product

Abstract

The application relates to the technical field of data processing, and discloses a multi-layer cognitive corpus construction method, equipment, medium and product. The method comprises the following steps: determining the state features of each modality candidate segment according to original multi-source task process data; determining candidate segments meeting a trigger condition as standard events based on the state features to form a unified event sequence; determining a segment division boundary of the unified event sequence according to dynamic mutation features of the unified event sequence in a preset dimension to generate a plurality of context-stable cognitive segments; extracting multi-dimensional features for each cognitive segment to obtain a segment-level basic feature vector; constructing a shared feature matrix according to the segment-level basic feature vector; and realizing the construction of a multi-layer cognitive corpus based on the shared feature matrix. The application can convert discrete multi-source task process data into a multi-layer structured cognitive corpus with context stability and multi-dimensional feature support, thereby improving the explainability and traceability of deep cognitive behavior.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, device, medium and product for constructing a multi-layer cognitive corpus. Background Technology

[0002] With the rapid development of large-scale language models and intelligent decision-making systems, deep behavioral modeling and cognitive deconstruction for complex multi-source tasks have become a core technological direction. Among these, multi-layered cognitive corpora refer to high-value structured data sets with multi-level cognitive structures, obtained by segmenting and reconstructing multimodal interaction data generated by humans or intelligent agents during the execution of specific tasks according to different cognitive depths (such as knowledge memory retrieval, logical deduction, causal alignment, and reflective constraints). Such corpora possess extremely high technical analytical value in constructing cognitive graphs for complex scenarios, evaluating cross-task inertial behavior, and conducting advanced compliance and risk control audits.

[0003] In related technologies, the recording and utilization of multi-source task process data typically employs rule-based log interception or direct data storage schemes for specific modal channels. Specifically, this involves capturing behavioral operation data such as mouse clicks and keyboard inputs through system embedding, or acquiring speech perception data through external audio acquisition devices. These heterogeneous streams of data from different channels are then archived directly into a database along a unified timeline, presented as independent physical tables, long text files, or audio files. When subsequent analysis is required, manual or shallow classifiers perform keyword retrieval for specific business tags or simple statistical classification on these relatively isolated data files.

[0004] However, the inventors have discovered at least the following technical problems in the related technologies: the corpus constructed in the related technologies remains in a shallow, discrete, and chronological form, and cannot objectively represent the deep cognitive behaviors that are intertwined at multiple levels in advanced human activities. Summary of the Invention

[0005] One objective of this application is to provide a method, device, medium, and product for constructing multi-layered cognitive corpora, at least to solve the technical problem that the corpora constructed in related technologies cannot objectively represent the multi-layered, intertwined, deep cognitive behaviors in advanced human activities.

[0006] To achieve the above objectives, some embodiments of this application provide the following aspects:

[0007] In a first aspect, some embodiments of this application provide a method for constructing a multi-layer cognitive corpus. The method includes: determining the state features of candidate segments of each modality based on original multi-source task process data; determining candidate segments that meet triggering conditions as standard events based on the state features to form a unified event sequence; determining the segment segmentation boundary of the unified event sequence based on the dynamic mutation features of the unified event sequence in a preset dimension to generate multiple context-stable cognitive segments; extracting multi-dimensional features for each cognitive segment to obtain a segment-level basic feature vector; constructing a shared feature matrix based on the segment-level basic feature vector; and constructing the multi-layer cognitive corpus based on the shared feature matrix.

[0008] Secondly, some embodiments of this application also provide an electronic device, the electronic device comprising: one or more processors; and a memory storing computer program instructions, which, when executed, cause the processor to perform the steps of the method described above.

[0009] Thirdly, some embodiments of this application also provide a computer-readable medium having computer program instructions stored thereon, which can be executed by a processor to implement the method described above.

[0010] Fourthly, some embodiments of this application also provide a computer program product, including computer program instructions that, when executed by a processor, implement the steps of the method described above.

[0011] Compared with related technologies, the solution provided in this application determines the state characteristics of each modality candidate segment based on the original multi-source task process data, and determines the candidate segments that meet the triggering conditions as standard events to form a unified event sequence based on the state characteristics. This can eliminate the inconsistencies and redundant conflicts of multi-source heterogeneous data at the time, space and business levels, and realize the standardized and unified processing of heterogeneous data on the same cognitive scale. By determining the segment segmentation boundary based on the dynamic mutation characteristics of the unified event sequence in a preset dimension, multiple context-stable cognitive segments are generated. Then, multi-dimensional features are extracted for each cognitive segment to obtain segment-level basic feature vectors and construct a shared feature matrix. Finally, multi-layer cognitive corpus is constructed based on the shared feature matrix, thereby transforming discrete process flow data into structured corpus with context stability and multi-dimensional feature support. This helps to solve the technical problem in related technologies that corpora can only record the appearance of mechanical behavior and cannot objectively represent the multi-level deep cognitive behavior in advanced human activities. It significantly improves the interpretability and traceability of cognitive corpus within the computer system. Attached Figure Description

[0012] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise specified, the figures in the drawings are not to be limited by scale.

[0013] Figure 1 An exemplary flowchart of a method for constructing a multi-layered cognitive corpus is provided for some embodiments;

[0014] Figure 2 A schematic diagram illustrating the unified event flow construction and cognitive segmentation process in a multi-layer cognitive corpus construction method provided for some embodiments;

[0015] Figure 3 An exemplary framework diagram of a nine-layer cognitive corpus in a multi-layer cognitive corpus construction method provided for some embodiments;

[0016] Figure 4 An exemplary schematic diagram of the inter-layer mapping relationship in a multi-layer cognitive corpus construction method provided in some embodiments;

[0017] Figure 5 An exemplary schematic diagram of a unified cognitive corpus structure in a multi-layer cognitive corpus construction method provided in some embodiments;

[0018] Figure 6 Another exemplary flowchart of a method for constructing a multi-layered cognitive corpus provided for some embodiments;

[0019] Figure 7 An exemplary structural diagram of an electronic device is provided for some embodiments. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] In this embodiment of the application, the collection, storage, use, processing, transmission, provision, and disclosure of user personal information comply with relevant laws and regulations and do not violate public order and good morals.

[0022] The following terms are used in this document.

[0023] Multi-layer Cognitive Corpus: refers to the structured results of human thinking experience organized according to different cognitive levels, including at least a portion of the following layers: knowledge layer, perception-representation layer, judgment layer, decision-making layer, association layer, physical cognition layer, metacognition layer, style and ability layer, and risk and normative constraint layer.

[0024] A cognitive segment is a relatively complete segment of a task process that is divided along a unified timeline. A cognitive segment must contain at least some information from the object, action, evidence, expression, feedback, or correction.

[0025] Inter-layer mapping refers to the referencing, supporting, triggering, constraining, or causal relationships between different cognitive layers. For example, the knowledge layer supports the judgment layer, the judgment layer triggers the decision layer, and the metacognitive layer regulates the decision layer.

[0026] Knowledge-layer Corpus Unit: refers to a structured knowledge unit extracted from cognitive fragments, including concepts, attributes, rules, hierarchies, relationships, and empirical conclusions that are actually invoked.

[0027] Perception-Representation Corpus Unit: refers to the mapping structural unit between external input information and internal cognitive representation.

[0028] Judgment-layer Corpus Unit: A structured unit that describes the evidence threshold, exclusion conditions, confidence boundaries, and flip boundaries when humans make a certain judgment.

[0029] Decision-layer Corpus Unit: A structured unit that describes the set of candidate actions, the set of excluded actions, the ordering of actions, the action locking conditions, and the backoff conditions.

[0030] Association-layer Corpus Unit: A structured unit that describes concept jumps, bridging cues, association types, and association paths.

[0031] Physical-Cognition-layer Corpus Unit: refers to a structured unit that describes the state, spatial relationships, actions, physical constraints, and predictions of physical consequences of an object.

[0032] Metacognition-layer Corpus Unit: refers to a structured unit that describes states and changes such as doubt, verification, self-correction, strategy switching, retaining uncertainty, and restoring certainty.

[0033] Style-and-Ability-layer Corpus Unit: A structured unit that describes an individual's stable thinking preferences, processing styles, and ability characteristics in multiple tasks.

[0034] Risk-and-Norm Constraint Corpus Unit: refers to a structured unit that describes how individuals identify and express control over the boundaries of risk, rules, ethics, and responsibility before answering, advising, judging, and making decisions.

[0035] Unified Cognitive Corpus Structure: refers to a unified data structure used to encapsulate the extraction results of different cognitive layers and the mapping relationships between them.

[0036] This embodiment relates to a method for constructing a multi-layered cognitive corpus. For example... Figure 1 As shown, the method may include the following steps:

[0037] Step S101: Determine the state characteristics of each modality candidate segment based on the original multi-source task process data;

[0038] Step S102: Based on the state characteristics, candidate segments that meet the triggering conditions are determined as standard events to form a unified event sequence;

[0039] Step S103: Based on the dynamic mutation characteristics of the unified event sequence in a preset dimension, determine the segmentation boundary of the unified event sequence to generate multiple context-stable cognitive segments.

[0040] Step S104: Extract multi-dimensional features for each cognitive segment to obtain a segment-level basic feature vector;

[0041] Step S105: Construct a shared feature matrix based on the fragment-level basic feature vectors;

[0042] Step S106: Based on the shared feature matrix, construct the multi-layer cognitive corpus.

[0043] The following provides a detailed explanation of each of the above steps:

[0044] In this embodiment, the original multi-source task process data refers to the original heterogeneous stream data collected from multiple different data sources and during the execution of multiple tasks. After receiving the multi-source task process data, the original multi-source task process data can be divided into modal channels at the data level. For each candidate segment within the divided modal channel, a feature extraction algorithm can be used to capture the real-time features of the candidate segment in terms of time, space, or action behavior, thereby determining the state features of each modal candidate segment.

[0045] After determining the state characteristics of each modality candidate segment, preset rules can be used to judge these characteristics. When the state characteristics of a candidate segment meet the triggering conditions, the candidate segment that meets the triggering conditions can be identified as a standard event, thereby standardizing the data format of the candidate segments. After determining the standard events, all generated standard events can be associated and arranged to eliminate redundancy or conflicts between different modalities, forming a unified event sequence aligned with multimodalities.

[0046] The unified event sequence is presented as continuous data on a timeline or business line. To obtain data segments with local semantic coherence, the progress of the unified event sequence can be evaluated from a preset dimension. When a dynamic mutation feature of the unified event sequence in a preset dimension is detected to reach a segmentation threshold, a segmentation boundary can be determined at the dynamic mutation location. After determining the segmentation boundary, data can be extracted from the unified event sequence based on the segmentation boundary to generate multiple context-stable cognitive segments.

[0047] Each generated cognitive segment can be considered an independent, context-stable unit. By extracting multi-dimensional textual features, behavioral features, and environmental state features within the cognitive segment, and fusing and vectorizing these multi-dimensional features, a segment-level basic feature vector is obtained for each cognitive segment. This segment-level basic feature vector represents the underlying multi-dimensional feature attributes of the cognitive segment.

[0048] After obtaining the fragment-level basic feature vectors of each cognitive segment, all fragment-level basic feature vectors can be aggregated. By combining the temporal correlation and contextual logical dependencies between cognitive segments, all fragment-level basic feature vectors are aligned and integrated in the feature space to construct a high-dimensional shared feature matrix. This shared feature matrix serves as a feature data pool, storing the underlying features throughout the entire task lifecycle, facilitating unified data retrieval during subsequent multi-layer cognitive corpus unit extraction.

[0049] Furthermore, after constructing the shared feature matrix, a multi-layer cognitive mapping mechanism can be used to extract cognitive corpus units of different cognitive dimensions from the shared feature matrix, and the cognitive corpus units can be structured to achieve the construction of multi-layer cognitive corpus based on the shared feature matrix.

[0050] The multi-layer cognitive corpus construction method provided in this embodiment can be applied to, but is not limited to: multi-layer cognitive corpus construction for industrial troubleshooting scenarios, multi-layer cognitive corpus construction for medical diagnosis experience, multi-layer cognitive corpus construction for educational tutoring scenarios, and multi-layer cognitive corpus construction for model training basic data construction. This embodiment does not make specific limitations on these applications.

[0051] It is not difficult to see that, compared with related technologies, the solution provided in this application, by determining the state characteristics of each modality candidate segment based on the original multi-source task process data, and determining the candidate segments that meet the triggering conditions as standard events based on the state characteristics to form a unified event sequence, can eliminate the inconsistencies and redundant conflicts of multi-source heterogeneous data at the time, space and business levels, and realize the standardized and unified processing of heterogeneous data on the same cognitive scale; by determining the segment segmentation boundary based on the dynamic mutation characteristics of the unified event sequence in a preset dimension to generate multiple context-stable cognitive segments, and then extracting multi-dimensional features for each cognitive segment to obtain segment-level basic feature vectors and constructing a shared feature matrix, and finally realizing the construction of multi-layer cognitive corpus based on the shared feature matrix, thereby transforming discrete process flow data into structured corpus with context stability and multi-dimensional feature support, which helps to solve the technical problem in related technologies that corpora can only record the appearance of mechanical behavior and cannot objectively represent multi-level deep cognitive behavior in advanced human activities, and significantly improves the interpretability and traceability of cognitive corpus within the computer system.

[0052] Optionally, in some embodiments, the original multi-source task process data includes at least two of the following: audio perception data, text interaction data, behavioral operation data, environmental state data, and verification feedback data.

[0053] For example, raw multi-source task process data may include multimodal data collected through various hardware devices or underlying interfaces. For instance: audio perception data may include voice data collected by a microphone; text interaction data may include text interaction records of keyboard input, such as text input and editing data; behavioral operation data may include user mouse clicks or shortcut key triggered behaviors captured by application system tracking points, such as operation logs and search records; environmental status data may include system resource utilization indicators or process status data, such as object status and scene data; and verification feedback data may include abnormal prompts or pop-up response results, such as feedback, risk warnings, and verification results.

[0054] For example, the raw multi-source task process data can be represented as follows:

[0055] ;

[0056] in, This represents the original multi-source task process data; This represents the audio perception data. This represents the text interaction data. This represents the behavioral operation data. This represents the environmental state data. This refers to the verification feedback data.

[0057] Optionally, in some embodiments, the step of determining the state characteristics of each modality candidate segment based on the original multi-source task process data, i.e., step S101, may include:

[0058] If the original multi-source task process data includes the audio perception data, then based on the audio perception data, extract the state features of the sentence-level segments as the candidate segments;

[0059] If the original multi-source task process data includes the text interaction data, then based on the text interaction data, extract the state features of the edited segment as the candidate segment;

[0060] If the original multi-source task process data includes the behavioral operation data, then based on the behavioral operation data, extract the state features of the atomic operation fragments as the candidate fragments;

[0061] If the original multi-source task process data includes the environmental state data, then based on the environmental state data, extract the state features of the state change segments as the candidate segments;

[0062] If the original multi-source task process data includes the verification feedback data, then the state features of the feedback fragments as candidate fragments are extracted based on the verification feedback data.

[0063] In other words, when determining the state characteristics of candidate segments for each modality, appropriate segment granularity segmentation and feature extraction can be adopted for different data modalities.

[0064] When the original multi-source task process data includes audio perception data, the audio perception data can be segmented within a modality, such as segmenting it into sentence-level segments based on silence intervals and semantic pauses as candidate segments, and extracting the volume fluctuations or pitch frequencies of the audio perception data as state features of the sentence-level segments.

[0065] When the original multi-source task process data includes text interaction data, the text interaction data can be segmented within a modality, such as by adding, deleting, replacing, or rewriting to form editing segments as candidate segments, and the input speed or text length of the text interaction data can be extracted as the state features of the editing segments.

[0066] When the original multi-source task process data includes behavioral operation data, the behavioral operation data can be segmented within a modality, such as by clicking, dragging, selecting, submitting, undoing, and rolling back, into atomic operation segments as candidate segments, and the operation time or control identifier of the behavioral operation data can be extracted as the state features of the atomic operation segments.

[0067] When the original multi-source task process data includes environmental state data, the environmental state data can be segmented within a modality, such as segmenting it into state change segments as candidate segments based on changes in object state, location, or relationship, and extracting the load change amount of the environmental state data as the state feature of the state change segment.

[0068] When the original multi-source task process data includes verification feedback data, the verification feedback data can be segmented within a modality, such as by verification, error message, external confirmation, and risk message, into feedback segments as candidate segments, and the confirmation delay time or response result of the verification feedback data can be extracted as the state features of the feedback segments.

[0069] It is not difficult to see that in the embodiments of this application, by specifically limiting the original multi-source task process data to at least two of the following: audio perception data, text interaction data, behavioral operation data, environmental state data, and verification feedback data, and correspondingly dividing them into sentence-level segments, editing segments, atomic operation segments, state change segments, and feedback segments for feature extraction, the determined state features are matched with the physical behavioral characteristics of the corresponding data modality. This can reduce the noise interference of heterogeneous data streams at the physical source, thereby providing more accurate multi-dimensional complementary feature inputs from the source.

[0070] Optionally, in some embodiments, the step of determining candidate fragments that meet the triggering conditions as standard events based on the state characteristics to form a unified event sequence may include:

[0071] Step S1021: Using the event triggering functions corresponding to the preset event triggering rules, the state features of each modal candidate segment are determined respectively;

[0072] Step S1022: When a certain state feature satisfies the corresponding event triggering function, the candidate segment to which the state feature belongs is converted into a standard event;

[0073] Step S1023: Arrange all the generated standard events in chronological order according to their timestamps, and divide the arranged standard event stream into local alignment windows;

[0074] In step S1024, within the local alignment window, if any two standard events meet the set time difference condition, the content relationship and object overlap relationship between the two standard events are determined; when the two standard events simultaneously meet the content similarity condition and the object overlap condition, the two standard events are repeatedly merged; when the two standard events meet the object overlap condition, and their content relationship indicates opposite operation types, state change directions, or semantic conclusions, the two standard events are conflict-splitting processed to generate the unified event sequence of multimodal alignment.

[0075] Optionally, in some embodiments, the preset event triggering rules include at least one of the following: voice pause rules, text editing behavior determination rules, atomic operation recognition rules, object state change rules, and verification feedback triggering rules.

[0076] In other words, when determining candidate segments that meet the triggering conditions as standard events, event triggering functions corresponding to each event triggering rule can be pre-established. Specifically, the speech pause rule can correspond to a silence duration determination function, the text editing behavior determination rule can correspond to a specific key press determination function, the atomic operation recognition rule can correspond to an operation combination matching function, the object state change rule can correspond to a configuration item change determination function, and the verification feedback triggering rule can correspond to an anomaly interception function.

[0077] For example, the event triggering function can be expressed as the following formula:

[0078] ;

[0079] in, This represents the event triggering determination function, whose output can be used to determine whether a candidate fragment is converted into a standard event; This refers to a specific data modality channel contained in the original multi-source task process data, such as audio perception data modality, text interaction data modality, or behavioral operation data modality; This indicates the time series index number of a candidate segment under a specific data modality channel; Indicates in the data modal channel Below, the time series index number is Candidate fragments; Indicates the data modal channel The established trigger scoring function; Indicates candidate fragments Feature significance score; Indicates the data modal channel Pre-set trigger threshold; This indicates an indicator function that outputs 1 if the condition within the parentheses is true and 0 if it is false.

[0080] The state features of each modality candidate segment can be input into the event triggering function for feature determination. When a certain state feature makes the trigger score of the corresponding modality greater than or equal to the set trigger threshold, i.e., the indicator function output is 1, the candidate segment is determined to meet the cognitive-related triggering conditions.

[0081] When the state features of each modality candidate fragment satisfy the corresponding event trigger function, a data packaging transformation can be triggered to standardize the data format of the candidate fragments, converting the candidate fragments that satisfy the cognitive-related trigger conditions into standard events. The standard events generated by the transformation can be expressed as tuples with standard data formats, specifically as the following structured data representation:

[0082] ;

[0083] in, Indicates the first A standard event; Indicates the first The absolute timestamps corresponding to each standard event can be used for precise temporal arrangement in a standard event stream; Indicates the first The data source modal channel for each standard event is used to identify the physical device or underlying interface from which the standard event originates. Indicates the first The business operation type of each standard event is used to classify and identify the nature of the behavior; Indicates the first The specific textual or physical content carried by a standard event; Indicates the first The object identifier that a standard event operates on, functions in, or is associated with; Indicates the first The context index number of a standard event is used to track the preceding and following contexts of the standard event in the task flow; Indicates the first The significance score of each standard event can be directly mapped and determined from the result calculated by the trigger scoring function; Indicates the first Each standard event points to a reference pointer to the original candidate fragment, used to preserve the link for tracing back to the original stream data; This represents a tuple data structure.

[0084] After generating the standard event tuple, all generated standard events can be arranged based on absolute timestamps, and local alignment windows can be divided into the arranged standard event stream. Within the local alignment window, it can first be checked whether any two standard events meet the set time difference condition; for two standard events that meet the time difference condition, the content relationship and object overlap relationship between them can then be determined.

[0085] For example, for any two standard events and ,in, This indicates that within the same local alignment window, it is related to standard events. The first step in comparison and judgment A standard event, if it meets the following time difference condition formula:

[0086] ;

[0087] Then standard events can be With standard events The elements are grouped into the same local alignment window. Represents standard events The corresponding absolute timestamp, Represents standard events The corresponding absolute timestamp; This represents a pre-set time difference threshold used to determine the local alignment window.

[0088] For example, the aforementioned time difference threshold The time can be between 0.2 seconds and 1.0 seconds.

[0089] Optionally, within the same local alignment window, a dual-condition determination can be made for the semantic content and associated objects of standard events. If two standard events simultaneously satisfy the following content similarity condition formula and object overlap condition formula:

[0090] ;

[0091] ;

[0092] in, This represents a predefined semantic content similarity calculation function, which can be used to quantify the degree of similarity between the text or behavioral content carried by two standard events; Represents standard events The specific text or behavioral content carried; Represents standard events The specific text or behavioral content carried; This represents a preset content similarity threshold, which serves as a critical value for determining whether two standard events are logically similar.

[0093] This represents a preset function for calculating object overlap or related object consistency, which can be used to determine whether two standard events have the same target. Represents standard events The identifier of the object being operated on or associated with; Represents standard events The identifier of the object being operated on or associated with; This represents the preset object overlap threshold, which serves as the critical value for determining whether two standard events act on the same target.

[0094] When two standard events satisfy the time difference condition, and simultaneously satisfy the content similarity condition and the object overlap condition, the two standard events can be determined to be duplicate events or logically convergent events for the same object, and a duplicate merging process is performed on the two standard events. The duplicate merging process may include retaining one of the standard events and merging the source modality, original candidate fragment pointer, saliency score, or other non-duplicate fields from the other standard event into the retained standard event.

[0095] When two standard events satisfy the time difference condition and the object overlap condition, but their content relationship indicates opposite operation types, state change directions, or semantic conclusions, it can be determined that the two standard events have a logical conflict. For example, one standard event indicates that an enable operation is performed on the target object, and the other standard event indicates that a deactivation operation is performed on the same target object; or, one standard event indicates that the state of the target object increases, and the other standard event indicates that the state of the target object decreases. In this case, conflict splitting processing can be performed on the two standard events, that is, the two standard events are not merged, but are retained as independent standard events, and their temporal order relationship and their respective source modal information are preserved.

[0096] When two standard events meet the time difference condition but not the object overlap condition, it can be determined that they act on different objects, and the two standard events are retained respectively without performing duplicate merging or conflict splitting.

[0097] After the duplicate merging or conflict splitting processes are completed within the local alignment window, a set of aligned and deduplicated standard events can be output to generate a unified event sequence for multimodal alignment. The final output of the unified event sequence can be represented as the following set:

[0098] ;

[0099] in, Represents a unified sequence of events aligned to multimodalities; This represents the first standard event arranged in chronological order. Represents the first event in the unified event sequence. A standard event; Represents the final generated unified event sequence set The total number of standard events after deduplication and alignment included in the document; Represents a unified event sequence The last standard event in the sequence of absolute timestamps.

[0100] It is not difficult to see that in the embodiments of this application, by using the event triggering functions corresponding to the preset event triggering rules to determine the state features, and converting the candidate fragments that satisfy the functions into standard events, the raw streaming data of different modalities can be uniformly converted into a standard event tuple structure, thereby solving the problem of inconsistent granularity and inconsistent fields caused by different collection sources in multi-source heterogeneous data. Furthermore, within the local alignment window divided based on the time difference threshold, according to the time difference condition, content similarity condition, object overlap condition, and content conflict relationship, duplicate or logically convergent standard events are subjected to duplicate merging processing, and standard events that act on the same object but have opposite content are subjected to conflict splitting processing. This can reduce duplicate events caused by synchronization errors of different modal devices and avoid the incorrect merging of conflicting events, thereby forming a logically consistent and temporally clear multimodal aligned unified event sequence. Through time alignment, duplicate merging, and conflict splitting, the interference of multimodal event redundancy and semantic conflict on subsequent multi-layer cognitive corpus extraction can be reduced, thereby improving the overall data accuracy of the unified event sequence.

[0101] Optionally, in some embodiments, the preset dimension includes at least one of the following: time interval dimension, semantic content dimension, task stage dimension, and feedback or risk status dimension.

[0102] Specifically, the preset dimensions are analytical dimensions used to measure whether the continuity of a unified event sequence data is broken. Among them, the time interval dimension is used to analyze the time span between adjacent standard events; the semantic content dimension is used to analyze the topic switching represented by text or behavior; the task stage dimension is used to analyze the crossing of main process nodes; and the feedback or risk status dimension is used to analyze whether abnormal status signals or verification instructions occur.

[0103] Optionally, in some embodiments, determining the segmentation boundaries of the unified event sequence based on the dynamic mutation characteristics of the unified event sequence in a preset dimension to generate multiple context-stable cognitive segments, i.e., step S103 may include:

[0104] Step S1031: Extract the feature signals corresponding to each dimension included in the preset dimension;

[0105] Step S1032: Calculate the comprehensive boundary determination value based on the feature signal;

[0106] Step S1033: When the comprehensive boundary determination value indicates that at least one of the preset dimensions has undergone a dynamic mutation, the corresponding event position is determined as the segmentation boundary of the unified event sequence.

[0107] Step S1034: Based on the segmentation boundaries, extract the sequence data between two adjacent segmentation boundaries in the unified event sequence to generate multiple context-stable cognitive segments.

[0108] In the specific implementation process, the process can proceed along a unified event sequence, extracting the feature signals corresponding to each dimension included in the preset dimensions. The feature signals may include time interval signals, semantic similarity drop signals, task stage transition signals, or risk state sudden increase signals.

[0109] To accurately identify logical breaks in the cognitive flow, a comprehensive boundary judgment value can be calculated using a boundary judgment function formula. Specifically, the boundary judgment function formula can be expressed as follows:

[0110] ;

[0111] in, Indicates the first The comprehensive boundary determination value at each event location is used to determine whether to establish a segmentation boundary at the current location. This represents the event position index in a unified event sequence; This represents a characteristic signal indicating the change in time interval between adjacent standard events; This represents a pre-defined threshold for time variation in the time interval dimension; Semantic abrupt change value feature signal representing text content or behavioral intent; This represents a pre-defined semantic mutation threshold for the semantic content dimension; This represents the characteristic signal of the task stage change value of the main process node; This represents the pre-defined threshold for stage changes in a task's stage dimension; This indicates a feedback or risk change value characteristic signal that represents an anomaly prompt or verification instruction; This indicates a pre-set threshold for risk changes based on feedback or risk status dimensions. Represents the logical OR operator; This indicates an indicator function that outputs 1 if either condition is true, and 0 otherwise.

[0112] The feature signals extracted from the unified event sequence can be input into the boundary judgment function formula mentioned above. When the comprehensive boundary judgment value... When =1, the current position can be determined as the segmentation boundary of the unified event sequence. By dynamically determining the segmentation boundary by combining boundary signals from multiple dimensions, the cognitive aliasing defects caused by mechanical segmentation with a fixed window can be avoided.

[0113] After determining the segmentation boundaries, the sequence data between two adjacent segmentation boundaries in a unified event sequence can be truncated according to these boundaries to generate multiple context-stable cognitive segments. The final set of truncated cognitive segments can be represented as follows:

[0114] ;

[0115] in, This represents the final set of cognitive fragments output. This represents the first cognitive segment unit in the set of cognitive segments; This indicates the last cognitive segment; This represents the total number of cognitive fragments generated in the end.

[0116] For the set of cognitive segments Any specific cognitive segment Its internal data structure can be expressed as a data structure with a stable context:

[0117] ;

[0118] in, Indicates the time series index number is Cognitive segments; This represents a local alignment standard event subsequence belonging to the cognitive segment unit; This represents the set of core operational objects involved within this cognitive segment unit; This represents the space of candidate actions available within the current cognitive segment unit context; This indicates the verification feedback status record triggered during the execution of this cognitive segment unit; This indicates the absolute time span of the life cycle of the cognitive segment unit.

[0119] Figure 2 This is a schematic diagram of the unified event flow and cognitive segmentation process (corresponding to the unified event sequence formation and cognitive segmentation process in this application) in an embodiment of the present application.

[0120] like Figure 2As shown in the attached figures, the raw data stream (corresponding to the raw multi-source task process data in this application) may include at least two of the following: speech (corresponding to the audio perception data in this application), text (corresponding to the text interaction data in this application), logs (corresponding to the behavioral operation data in this application), and scene / feedback (corresponding to the environmental state data and / or verification feedback data in this application).

[0121] Based on the modal data, intra-modal segmentation can be performed to obtain the atomic event fragments (e.g., statement-level, edit-level, and operation-level) shown in the attached figures (corresponding to the modal candidate fragments in this application). Specifically, statement-level atomic event fragments can correspond to sentence-level fragments, edit-level atomic event fragments can correspond to edit fragments, and operation-level atomic event fragments can correspond to atomic operation fragments. Furthermore, each modal candidate fragment may also include state change fragments obtained from environmental state data and feedback fragments obtained from verification feedback data.

[0122] Using event triggering functions corresponding to preset event triggering rules, the state characteristics of each modal candidate segment are determined, and candidate segments that meet the triggering conditions are converted into standard events. The candidate events in the attached figure (corresponding to the standard events obtained by triggering determination and conversion in this application and participating in the construction of the unified event sequence) can be arranged in chronological order according to absolute timestamps.

[0123] The time alignment and conflict resolution shown in the attached figures are performed on the arranged standard events (corresponding to the time arrangement, duplicate merging, or conflict splitting of standard events in this application). The figures show... The time difference threshold used when dividing the local alignment window can be between 0.2 seconds and 1.0 seconds. Within the same local alignment window, standard events that meet the time difference condition, content similarity condition, and object overlap condition can be merged repeatedly; standard events that meet the time difference condition and object overlap condition but have opposite operation types, state change directions, or semantic conclusions can be split due to conflict, and the corresponding standard events can be retained separately.

[0124] After completing temporal alignment and conflict resolution, a unified event sequence aligned to multiple modalities can be formed. Dynamic mutation features along a preset dimension are detected along this unified event sequence, wherein, as shown in the attached figure… (The time interval change characteristic signal in this application being greater than the time change threshold) indicates a dynamic abrupt change in the time interval between adjacent standard events; the attached figures... (The semantic content change feature signal in this application is greater than the semantic mutation threshold) indicates that the text content or behavioral intent between adjacent standard events has undergone dynamic mutation.

[0125] When at least one of the dimensions of time interval, semantic content, task stage, or feedback or risk state undergoes a dynamic mutation, the segmentation boundary can be determined at the corresponding event location, and the unified event sequence can be truncated according to the segmentation boundaries of two adjacent segments, thereby generating cognitive segment 1, cognitive segment 2, and cognitive segment 3 (corresponding to multiple context-stable cognitive segments in this application) as shown in the attached figure. Figure 2 The three cognitive segments shown are for illustrative purposes only. The actual number of cognitive segments generated can be determined based on the number of segmentation boundaries detected in the unified event sequence.

[0126] It is not difficult to see that in the embodiments of this application, by refining the preset dimensions into at least one of the time interval dimension, semantic content dimension, task stage dimension, and feedback or risk state dimension, and using an indicator function to jointly determine whether the feature signals of each dimension are greater than the corresponding dimension threshold, and determining the segmentation boundary when at least one dimension undergoes a dynamic mutation, new cognitive segments can be adaptively segmented when any change in the time interval, semantic content, task stage, or feedback / risk state exceeds the threshold. This segmentation limitation utilizes the quantization logic or fusion of multi-dimensional boundary feature signals, overcoming the limitations of fixed time steps or single window statistical indicators in identifying contextual breaks. It can accurately capture abrupt breaks in the logical shift of the data stream, avoiding cognitive aliasing caused by fixed window segmentation, thereby enabling each extracted cognitive segment to have more objective stability and logical completeness in the internal context of the segment.

[0127] Optionally, in some embodiments, the construction of the multi-layer cognitive corpus based on the shared feature matrix, i.e., step S106, may include:

[0128] Step S1061: Establish scoring functions and judgment conditions for the multiple cognitive layers constituting the multi-layer cognitive corpus respectively, and extract multi-layer cognitive corpus units based on each cognitive segment and the feature data corresponding to the cognitive segment in the shared feature matrix.

[0129] Step S1062: The multi-layer cognitive corpus units are structured to realize the construction of the multi-layer cognitive corpus.

[0130] Specifically, independent scoring functions and decision conditions can be deployed for multiple different cognitive dimensions. Using these independent scoring functions and decision conditions, multi-layered cognitive corpus units can be extracted from each cognitive segment and its corresponding shared feature matrix. After extracting the multi-layered cognitive corpus units, the data format of these units can be converted and standardized using the independent structured output units corresponding to each cognitive layer, thereby generating layer units that conform to a preset data structure, thus realizing the construction of the multi-layered cognitive corpus.

[0131] For example, for each generated cognitive segment, the cognitive segment can be treated as an independent context-stable unit for segment-level basic feature extraction. If the multi-layered cognitive corpus unit has nine layers, to meet the common input requirements of nine-layer cognitive extraction, object features, evidence features, expression features, action features, association features, physical relationship features, metacognitive features, risk features, and feedback features can be uniformly extracted for each cognitive segment. The resulting segment-level basic feature vector can be expressed as the following dense vector formula:

[0132] ;

[0133] in, Indicates the time series index number is The segment-level basic feature vector corresponding to the cognitive segment unit; This represents the object feature components extracted within the cognitive segment unit; This represents the evidence feature components extracted within the cognitive segment unit; This represents the expressive feature components extracted from within the cognitive segment unit; This represents the action feature components extracted within the cognitive segment unit; This represents the associative feature components extracted within the cognitive segment unit; This represents the physical relationship feature components extracted within the cognitive segment unit; This represents the metacognitive feature components extracted within the cognitive segment unit; This represents the risk feature components extracted within the cognitive segment unit; This represents the feedback feature component extracted from within the cognitive segment unit.

[0134] Furthermore, in order to accurately characterize the deep state of cognitive segments at the interactive expression level, the aforementioned segment-level basic feature vectors The expressive feature components contained therein It can be explicitly defined as the following parametric subvector formula:

[0135] ;

[0136] in, The conclusion expression strength parameter is used to quantify the significance of logical assertions or affirmative statements within a cognitive segment; This parameter represents the intensity of uncertain expression and is used to quantify the frequency of occurrence of ambiguous or probabilistic inference words within a cognitive segment. The parameter representing the intensity of pauses or hesitations is used to quantify the non-functional duration of speech pauses or operation intervals in a temporal sequence. This represents the correction intensity parameter, used to quantify the frequency of text being deleted, rewritten, or rolled back. This parameter represents the change in granularity, used to quantify the step difference in the level of detail of data fields between successive operations. This parameter represents the verification intensity, used to quantify the frequency of query, verification, external confirmation, or validation operations within a cognitive segment. The verification intensity parameter can be determined based on validation operation records, external confirmation records, or validation results in the verification feedback data, and can be normalized according to the total number of events or their duration within the cognitive segment.

[0137] In this embodiment, the feature signals used by the cognitive layer scoring functions, judgment functions, and inter-layer relationship scoring functions described below can all be obtained from the original multi-source task process data, standard events, cognitive segments, segment-level basic feature vectors, or shared feature matrices. Specifically, frequency-based feature signals can be determined based on the number of times the corresponding event or operation occurs within a cognitive segment, and can be normalized according to the total number of events contained in the cognitive segment or the duration of the cognitive segment; time-based feature signals can be calculated based on the timestamps of the corresponding standard events; object change-based feature signals can be determined based on whether the object identifiers in adjacent standard events have changed; text semantic-based feature signals can be determined based on the cosine similarity, Euclidean distance, or other preset similarity calculation methods between text feature vectors; state difference-based feature signals can be determined based on the difference between corresponding state fields, whether the categories are consistent, or whether the relationships are consistent; configuration features such as risk signals, specification sources, business importance, and responsibility boundaries can be read from the rule base, configuration table, or permission information corresponding to the current business scenario.

[0138] Before being input into the corresponding scoring function, each feature signal can be normalized based on the maximum and minimum values ​​in historical samples, or standardized according to a preset statistical interval. The weight coefficients and decision thresholds in each scoring function are pre-configured model parameters. They can be determined by searching for parameters on the validation set based on labeled historical task process data, targeting classification accuracy, recall, or a comprehensive evaluation index. Alternatively, they can be determined based on the quantiles of the historical sample score distribution. In the absence of labeled samples, equal weights can be used as initial weights, which can then be adjusted by business personnel based on actual operating results. The determined weight coefficients and decision thresholds are stored in a parameter configuration table associated with the corresponding cognitive layer and are read from this parameter configuration table when the corresponding scoring function is executed.

[0139] For a scoring function that contains only positive weighting terms, after normalizing each feature signal to the 0 to 1 interval, each weight coefficient can be set to a non-negative number and the sum of the weight coefficients can be 1, and the corresponding threshold can be set within the 0 to 1 interval. For a scoring function that contains both positive weighting terms and negative deduction terms, the corresponding threshold can be set within the output range of the scoring function defined by each weight coefficient.

[0140] After obtaining the segment-level basic feature vectors for each cognitive segment, the object feature component, evidence feature component, expression feature component, action feature component, association feature component, physical relationship feature component, metacognitive feature component, risk feature component, and feedback feature component in each segment-level basic feature vector can be concatenated according to a pre-defined feature component order. For feature components with different dimensions, they can be unified to the preset dimension through truncation, zero-padding, or feature mapping. For feature components missing due to the absence of the corresponding data modality in the original multi-source task process data, preset default values ​​can be filled in, and the missing status of the feature component can be recorded through the corresponding missing flag.

[0141] The segment-level basic feature vectors corresponding to each cognitive segment are arranged row by row according to the temporal order of the cognitive segments, so that the k-th row of the shared feature matrix corresponds to the cognitive segment with temporal index number k. Each column of the matrix corresponds to a pre-defined feature dimension, thereby constructing the shared feature matrix. Based on the shared feature matrix, the scoring function of each cognitive layer can read the row vector corresponding to the current cognitive segment, and when it is necessary to analyze adjacent segments or cross-task features, it can read the row vectors corresponding to multiple cognitive segments for calculation.

[0142] It is not difficult to see that in the embodiments of this application, by unifying the multi-source task process data into events and constructing cognitive fragments, not only are low-significance redundancy in multimodal data filtered out, but also a unified spatiotemporal physical boundary and logical anchor point are successfully established, effectively solving the technical defect that it is difficult to uniformly process and fuse original interactive data of different modalities and dimensions on the same cognitive scale.

[0143] Optionally, in some embodiments, the multi-layer cognitive corpus unit is specifically a nine-layer cognitive corpus unit; wherein, the nine-layer cognitive corpus unit includes: a knowledge layer corpus unit, a perception-representation layer corpus unit, a judgment layer corpus unit, a decision-making layer corpus unit, an association layer corpus unit, a physical cognition layer corpus unit, a metacognitive layer corpus unit, a style and ability layer corpus unit, and a risk and normative constraint layer corpus unit.

[0144] Specifically, such as Figure 3 As shown, the multi-layered cognitive corpus unit specifically consists of nine layers, including: the Knowledge Layer Corpus Unit (KCU) corresponding to the invocation of knowledge entities; the Perception-Representation Layer Corpus Unit (RCU) corresponding to the perceptual representation of environmental states; the Judgment Layer Corpus Unit (JCU) corresponding to the logical deduction process; the Decision Layer Corpus Unit (DCU) corresponding to the selection of behavioral paths; and the Association Layer Corpus Unit (ACU, corresponding to the attachment...). Figure 3 The "Association Layer (ACU)" in the diagram represents concept-jumping behavior; the Physical Cognition Layer Corpus Unit (PCU) corresponds to the alignment of physical causal relationships; the Metacognition Layer Corpus Unit (MCU) corresponds to confidence reflection; the Style and Capability Layer Corpus Unit (SCU) corresponds to cross-task style stability; and the Risk and Normative Constraint Layer Corpus Unit (QCU) corresponds to safety red lines or compliance constraints. Inter-layer mapping relationships exist between the cognitive layer corpus units, as shown in the attached diagram. These inter-layer mapping relationships include support relationships, triggering relationships, constraint relationships, and regulatory relationships (i.e., "support, trigger, constraint, and regulation" shown in the attached diagram), used to represent the logical synergy, evolutionary causal relationships, or output control mechanisms between different cognitive layer corpus units.

[0145] For example, the cognitive fragments can be used and the corresponding fragment-level basic feature vector Using the common input, and employing computational scoring functions and decision boundaries deployed independently for each cognitive level, nine cognitive corpus units of different cognitive levels are extracted from cognitive segments and their corresponding shared feature matrices.

[0146] It is not difficult to see that in the embodiments of this application, by constructing standardized corpus units for nine different cognitive layers of thinking, namely the knowledge layer, perception-representation layer, judgment layer, decision-making layer, association layer, physical cognition layer, metacognition layer, style and ability layer, and risk and normative constraint layer, a three-dimensional and in-depth deconstruction of the entire scenario and chain of human subjective dynamic thinking is achieved. This helps to solve the technical defects of existing technologies that have single corpus extraction methods and can only extract superficial behaviors or single cognitive aspects in isolation.

[0147] Optionally, in some embodiments, the step of establishing scoring functions and judgment conditions for each of the multiple cognitive layers constituting the multi-layered cognitive corpus, and extracting multi-layered cognitive corpus units based on each cognitive segment and the feature data corresponding to that cognitive segment in the shared feature matrix, may include:

[0148] The knowledge layer corpus units are determined based on the knowledge retrieval intensity of each knowledge unit and the set knowledge layer threshold;

[0149] The perception-representation layer corpus unit is determined based on the focus switching score and the set perception threshold; the focus switching score is determined based on at least one of object change features, gaze or trajectory offset features, semantic focus deviation features, and interface control switching features.

[0150] The judgment layer corpus unit is determined based on the judgment score, the set validity threshold, the set flip boundary, and the exclusion conditions; the judgment score is determined based on positive evidence and counter-evidence.

[0151] The decision-making layer corpus units are determined based on the decision compression rate; the decision compression rate is calculated based on candidate actions and excluded actions.

[0152] The association layer corpus unit is determined based on the association trigger score and the set association threshold; the association trigger score is determined based on concept jump events.

[0153] The physical consequence deviation is used to determine the physical cognition layer corpus unit; the physical consequence deviation is calculated based on the alignment of predicted consequences and actual consequences in the physical cognition elements.

[0154] The metacognitive layer corpus unit is determined based on the change in confidence boundaries; the change in confidence boundaries is determined based on expressive features and adjacent cognitive segments.

[0155] The style and capability layer corpus units are determined based on style stability; the style stability is determined based on cross-task features reflecting multi-task scenarios.

[0156] The risk and normative constraint layer corpus units are determined based on the corpus suppression score; the corpus suppression score is determined based on risk signals, normative sources, and responsibility boundaries.

[0157] Specifically, the quantification and extraction process of corpus units at each cognitive level is as follows:

[0158] For knowledge layer corpus units, access data of each knowledge unit within a cognitive segment can be statistically analyzed to calculate the knowledge call intensity, and the knowledge layer corpus unit is determined when the knowledge call intensity is greater than or equal to the set knowledge layer threshold.

[0159] For example, concepts, rules, attributes, levels, and experiential conclusions within the cognitive segment units can be identified. To achieve the merging and quantification of multi-dimensional features, a knowledge layer scoring function, i.e., a knowledge retrieval intensity scoring formula, can be used for quantification calculation. Specifically, the knowledge retrieval intensity scoring formula can be expressed as:

[0160] ;

[0161] in, This refers to a specific knowledge unit identified within the cognitive segment unit; Represents a specific unit of knowledge The corresponding knowledge retrieval intensity score; Represents a specific unit of knowledge Frequency characteristic signals that are accessed or invoked within the cognitive segment unit; Represents a specific unit of knowledge The depth-of-depth feature signals referenced or associated within the cognitive segment unit; Represents a specific unit of knowledge The corresponding business importance characteristic signal; Represents a specific unit of knowledge Frequency characteristic signals that are revoked, modified, or negated within the cognitive segment unit; This represents the pre-defined weighting constants and weighting coefficients for each dimension.

[0162] Specific knowledge units The characteristic signals of each element are input into the above knowledge retrieval strength scoring formula. The calculated knowledge retrieval strength score satisfies the following set judgment condition formula:

[0163] ;

[0164] In the above determination formula, This represents a pre-set knowledge layer threshold; when a specific knowledge unit... When the knowledge call intensity score is greater than or equal to the knowledge layer threshold, a specific knowledge unit is determined. Truly participating in the current thought process allows for the processing of specific knowledge units. It was included and identified as a knowledge-layer corpus unit.

[0165] The knowledge layer corpus units that are finally extracted and packaged can be expressed as a tuple data structure with a standardized data format:

[0166] ;

[0167] in, Indicates the time series index number is The knowledge layer corpus units corresponding to the cognitive fragment units; This represents the set of core concepts belonging to the corpus unit of this knowledge layer. This represents the set of business rules or logical theorems belonging to the corpus unit of this knowledge layer. A set of attribute features representing a specific concept or rule; This indicates the cognitive level or technical depth during the knowledge retrieval process. This indicates the basic uncertainty index value corresponding to the knowledge retrieval; This indicates the application boundaries or applicable scope conditions corresponding to the knowledge unit.

[0168] By identifying concepts, rules, attributes, levels, and experiential conclusions in the cognitive fragment units, and calculating scores based on a knowledge call intensity scoring formula that includes access frequency, citation depth, business importance, and retraction / correction frequency, specific knowledge units are encapsulated into a knowledge layer corpus unit tuple structure containing a core concept set, a business rule set, an attribute feature set, a cognitive level marker, a basic uncertainty index value, and application boundary conditions when the score is greater than or equal to a set knowledge layer threshold. This can accurately distinguish between knowledge that has only "appeared" in the interaction flow and knowledge that "truly participates in the current thought process," filtering out occasional background noise, thereby improving the physical effectiveness and behavioral representation ability of knowledge layer corpus units.

[0169] For the perception-representation layer corpus units, it is possible to identify external objects, attentional focus, and ignored objects involved within the cognitive segment unit, and determine whether the attentional focus has switched during the interaction process. To accurately quantify the spatial or perspective shift state, a perception layer scoring function, i.e., a focus switching score formula, can be used for quantitative calculation. Specifically, the focus switching score formula can be expressed as:

[0170] ;

[0171] in, This represents the attention focus switching score calculated within the cognitive segment unit; This indicates a change signal that represents a change in the physically associated object between consecutive operations; The offset characteristic signal represents the line-of-sight point or trajectory interaction in the sensing channel; This indicates the deviation value feature signal in the core descriptive theme or semantic focus of text interaction data; Physical characteristic signals indicating when a UI control element is clicked, selected, or hovered; This represents the pre-defined weighting constants used to balance the various focus detection dimensions.

[0172] The characteristic signals can be input into the above focus switching score formula. The calculated focus switching score satisfies the following set judgment condition formula:

[0173] ;

[0174] In the above determination formula, This represents a pre-set perception threshold, i.e., a focus switching determination threshold. When the focus switching score is greater than or equal to the perception threshold, it is determined that a significant shift in attention has occurred, thereby triggering the generation of a new internal representation mapping. The finally extracted and packaged perception-representation layer corpus units can be expressed as a tuple data structure with a standardized data format:

[0175] ;

[0176] in, Indicates the time series index number is The cognitive segment unit corresponds to the perceptual-representational layer corpus unit; This represents the set of external object information belonging to the internal structure of this perception-representation layer corpus unit; This represents the focus of attention or the set of highlighted features that are currently being considered in the thought process. This represents a set of background objects or noise information that are filtered or ignored in the current context. This represents the internal representation mapping relationship formed by the projection of external objects onto the internal cognitive space; These are feature parameters that indicate the trajectory of attention shifting between different physical objects or semantic topics.

[0177] When extracting units from the perception-representation layer corpus, the focus switching score is calculated based on the physical features of physical object changes, gaze shifts, deviations from the descriptive topic, and control switching. When the score is greater than or equal to the set perception threshold, a new internal representation mapping is adaptively generated. This can identify the dynamic shifts in the user's attention distribution and the filtering and selection features of external environmental information, thereby greatly improving the accuracy of the multimodal corpus in reconstructing the collaborative mapping of subjective cognitive states and the external physical world.

[0178] For each judgment-level corpus unit, the system can identify the judgment target, positive evidence, negative evidence, and exclusion conditions within the cognitive segment unit, and calculate a judgment score to determine whether the viewpoint is valid, pending confirmation, or reversed / cancelled during the judgment process. The judgment score is determined based on positive and negative evidence, and the exclusion conditions serve as additional verification criteria to determine whether the judgment result is allowed to be valid.

[0179] To construct the quantization calculation logic, a judgment layer scoring function, i.e., a judgment score formula, can be used for dynamic quantization calculation. Specifically, the judgment score formula can be expressed as follows:

[0180] ;

[0181] in, This represents the judgment score calculated within the cognitive segment unit; Indicates the identified first A characteristic signal of positive evidence supporting the current judgment objective; Indicates the first Preset credibility weight coefficients corresponding to the feature signals of each positive piece of evidence; This represents the total number of positive evidence feature signals identified. Indicates the identified first A characteristic signal of counter-evidence that refutes the current judgment target; Indicates the first Preset rebuttal weight coefficients corresponding to the feature signals of each counter-evidence; This represents the total number of characteristic signals of the identified counter-evidence.

[0182] The characteristic signals of various pieces of evidence can be input into the above judgment scoring formula. In the specific state determination, a dual dynamic critical boundary can be set, and the specific judgment logic is as follows:

[0183] When the calculated judgment score satisfies the following set judgment condition formula:

[0184] ;

[0185] In the above formula for the conditions for validity, This indicates the pre-set threshold for the condition to be met.

[0186] When the calculated judgment score satisfies the following judgment reversal condition formula:

[0187] ;

[0188] In the above formula for the flipping condition, No. The flip boundary used by each judgment layer corpus unit, and the flip boundary and the establishment threshold satisfy the following preset inequality constraint:

[0189] ;

[0190] Specifically, when the judgment score is greater than or equal to the validity threshold and there are no exclusion conditions that are currently met, the current judgment conclusion is determined to be valid, and the corresponding judgment validity state is generated.

[0191] When the judgment score is less than the flip boundary, and there is an established judgment for the same judgment target in the preceding cognitive segment, it is determined that the established judgment is flipped or revoked, and a corresponding judgment flip state is generated; if there is no established judgment for the same judgment target in the preceding cognitive segment, the current judgment state can be recorded as an invalid state.

[0192] When the judgment score is greater than or equal to the flip boundary and less than the establishment threshold, it is considered to be satisfied. When the current judgment state is pending confirmation, the judgment result is not output as either true or reversed, and the judgment is awaited for new evidence in subsequent cognitive segments to be judged again.

[0193] When the exclusion condition is met, even if the judgment score reaches the threshold for validity, the judgment result is not output. Instead, the corresponding exclusion condition is recorded, and the current judgment state is determined to be the exclusion state. The exclusion condition can be read from the business rules, restrictions, permission configurations, or risk rules corresponding to the current task. The condition is determined to be met by judging whether the object, state, or operation type defined by the exclusion condition matches the current cognitive segment.

[0194] The final extracted and packaged decision layer corpus units can be expressed as a tuple data structure with a standardized data format:

[0195] ;

[0196] in, Indicates the time series index number is The cognitive segment unit corresponds to the judgment layer corpus unit; This indicates the judgment result corresponding to the judgment layer corpus unit. The judgment result may include a judgment target identifier and a judgment state among: established, pending confirmation, flipped or canceled, excluded, and not established. This represents the set of positive evidence belonging to the corpus unit of this judgment layer; This represents the set of opposing evidence belonging to the corpus unit of this judgment layer; This represents the set of exclusion conditions or negation constraint boundaries that are active in the current context; This indicates the threshold value used by the current judgment layer corpus unit; Indicates the flip boundary used by the current decision layer corpus unit; This represents the overall confidence score based on a quantitative comparison of current positive and negative evidence.

[0197] In some embodiments, the overall confidence score can be determined based on the position of the judgment score within a preset value range. Specifically, the judgment score can be normalized to the interval between 0 and 1 according to the minimum and maximum possible scores of the corresponding scoring function, and the normalization result is determined as the overall confidence score. A higher overall confidence score indicates a higher degree of support between the current positive evidence and the negative evidence.

[0198] By identifying the judgment target, positive evidence, negative evidence, and exclusion conditions in cognitive segment units, and calculating scores based on a judgment scoring formula that includes dynamic weighting of positive evidence and dynamic deduction of negative evidence, the judgment state is determined as valid, pending confirmation, reversed or revoked, excluded, or invalid based on the validity threshold and reversal boundary interval of the judgment score and whether the exclusion conditions are met. The judgment information is encapsulated into a tuple structure of judgment layer corpus units containing the judgment result, a set of positive and negative evidence, exclusion conditions, dual thresholds, and a comprehensive confidence score. This not only objectively outputs the final judgment conclusion but also structurally expresses the boundary conditions, failure boundaries, and dynamic reversal mechanisms for the validity of the judgment conclusion. It preserves the evidence balancing trajectory of the judgment process when facing conflicting information, thereby improving the computability, traceability, and logical deduction depth of the judgment process within the computer system.

[0199] For decision-level corpus units, it is possible to identify initial candidate actions available within the cognitive segment unit and excluded actions that are not satisfied with the constraints. The initial candidate actions can be obtained from the task flow configuration, the set of operations allowed to be executed under the current business state, or the set of candidate operations generated by the model; the excluded actions can be determined by matching each initial candidate action with the current task constraints, permission constraints, risk constraints, or resource constraints.

[0200] The decision compression ratio can be determined based on the ratio of the number of excluded actions to the total number of initial candidate actions. A higher decision compression ratio indicates a greater degree of compression of the candidate action space after constraint processing. For each candidate action that is not excluded, an action evaluation score can be determined based on at least one of the following: the matching degree between the candidate action and the current task objective, execution cost, expected benefit, and risk level. The matching degree, execution cost, expected benefit, and risk level can be read or calculated from task configuration, historical execution records, risk rule base, or shared feature matrix.

[0201] The candidate action with the highest action evaluation score is determined as the highest priority action, and the candidate action with the second highest action evaluation score is determined as the second highest priority action. The priority difference between the two is calculated. When the decision compression rate reaches a preset compression rate threshold, and the priority difference reaches a preset priority difference threshold, the highest priority action is determined as the locked action, and a decision-level corpus unit is generated. The decision-level corpus unit includes at least an initial candidate action set, an excluded action set, a locked action, a decision compression rate, and an action evaluation score.

[0202] The compression ratio threshold and priority difference threshold can be determined statistically or verified based on the confirmed decision-locked samples in historical tasks, and stored in the decision layer parameter configuration table.

[0203] For the associative layer corpus units, we can detect concept jump events within cognitive segment units and identify bridging cues that connect different concepts to determine the associative activation state during the thinking and decision-making process. To construct a quantitative calculation logic, we can introduce associative jump distance and associative trigger score for multi-dimensional quantitative evaluation.

[0204] First, to quantify the logical span between the source concept and the target concept, the associative jump distance formula can be used for calculation. Specifically, the associative jump distance formula can be expressed as follows:

[0205] ;

[0206] in, Indicates the first The associative jump distance corresponding to each concept jump event; This represents the source concept that triggers the association. This indicates the target concept that the association points to; It represents the semantic similarity between the source concept and the target concept.

[0207] Furthermore, in order to comprehensively evaluate the activation intensity of associative activities, the associative layer scoring function, i.e., the associative triggering score formula, can be used for dynamic quantitative calculation. Specifically, the associative triggering score formula can be expressed as follows:

[0208] ;

[0209] in, This represents the associative trigger score calculated for the current cognitive segment unit; These are activating signals that indicate the presence of cues or clues in the external environment or interactive context. This represents the contrast and conflict characteristic signal between the source concept and the target concept. Analogous mapping feature signals representing cross-domain structural similarity; This represents the memory retrieval strength feature signal of historically related experiences in the current context; This represents the pre-set weighting constants used to balance the various association detection dimensions.

[0210] The various jump and clue characteristic signals can be input into the above formula. When specifically determining the state, the calculated association trigger score satisfies the following set determination condition formula:

[0211] ;

[0212] In the above determination formula, This represents a pre-set association threshold; when the association trigger score is greater than or equal to the association threshold, it is determined that the current concept jump constitutes a valid association chain, thereby triggering the generation of association layer corpus units. The finally extracted and packaged association layer corpus units can be expressed as a tuple data structure with a standardized data format:

[0213] ;

[0214] Indicates the time series index number is The cognitive fragment unit corresponds to the associative layer corpus unit; This represents the source concept corresponding to the corpus unit in the association layer; This indicates the target concept corresponding to the corpus unit in the association layer; A bridging cue component that serves as a physical or semantic link between the source concept and the target concept; This indicates the specific type of jump in the associative activity, such as analogy, causal association, or proximity association; This represents the calculated associative jump distance between the source concept and the target concept; This indicates the explicit confidence score corresponding to the associative chain; This represents the topological path parameters that record the dynamic evolution trajectory of association.

[0215] By detecting concept jump events in cognitive segments and identifying bridging cues, and calculating scores based on the associative jump distance formula determined by the semantic similarity between the source and target concepts, combined with the associative triggering score formula that includes cueing cues, contrast conflicts, analogy mappings, and memory retrieval strength, when the score is greater than or equal to a set associative threshold, the associative information is encapsulated into an associative layer corpus unit tuple structure containing the source concept, target concept, bridging cues, jump type, jump distance, confidence level, and topological path. This can explicitly explain the fundamental causes of associative activities, the logical bridging mediates they depend on, and the spatial span of thought crossings, effectively avoiding the content discontinuity problem caused by leaps in thought in historical interactive corpora, thereby improving the structuring degree and causal inference transparency of subjective associative paths within the computer system.

[0216] For physical cognitive layer corpus units, physical cognitive elements can be extracted and the predicted consequences can be aligned with the actual consequences. The deviation between the predicted and actual consequences can be calculated, and the physical cognitive layer corpus units can be determined based on the deviation.

[0217] Specifically, external objects, three-dimensional space, contact relationships, support relationships, action changes, and physical constraints involved within a cognitive segment unit can be extracted. The predicted consequences of changes in the physical environment are then aligned with the actual consequences to calculate the physical consequence deviation. To construct the quantitative calculation logic, the physical cognitive layer scoring function, i.e., the physical consequence deviation formula, can be used for dynamic quantitative calculation. The physical consequence deviation formula can be specifically expressed as:

[0218] ;

[0219] in, This represents the deviation in physical consequences calculated for the current cognitive segment unit; The characteristic signal representing the type difference component between the predicted consequences and the actual consequences; The characteristic signal representing the state difference component identified between the predicted consequences and the actual consequences; The feature signal representing the spatial or logical difference between the predicted consequences and the actual consequences; This represents the pre-defined weighting constants used to balance the various disparity dimensions.

[0220] The physical difference characteristic signals can be input into the above formula for physical consequence deviation for calculation. The calculated physical consequence deviation can characterize the degree of deviation between the predicted consequence and the actual consequence, and serve as a quantitative basis for determining the deviation state corresponding to the current physical cognitive element and constructing physical cognitive layer corpus units. Finally, the extracted and packaged physical cognitive layer corpus units can be expressed as a tuple data structure with a standardized data format:

[0221] ;

[0222] in, Indicates the time series index number is The physical cognitive layer corpus units corresponding to the cognitive segment units; This represents the basic physical cognitive element unit belonging to the corpus unit of this physical cognitive layer, which may specifically include the manipulated physical object, the spatial coordinates of the object, the surface contact state, and the spatial support constraint relationship. This represents the set of physical action features applied to the basic element unit of physical cognition; This represents the set of environmental physical constraints corresponding to the current physical situation. It represents a record of the predicted consequences of a physical action, estimated or anticipated in advance by subjective thinking. This represents the actual physical consequences returned by the underlying sensors or interactive interfaces after a physical action is performed. This represents the magnitude of the deviation between the calculated predicted consequences and the actual consequences in terms of physical consequences.

[0223] By extracting objects, space, contact, support, actions, and physical constraints within cognitive fragment units, and aligning predicted consequences with actual consequences using a formula for physical consequence deviation that includes a weighted sum of type differences, state differences, and relational differences, the physical cognitive information is ultimately encapsulated into a tuple structure of physical cognitive layer corpus units. This structure includes basic physical cognitive elements, a set of physical actions, environmental physical constraints, records of predicted consequences, records of actual consequences, and physical consequence deviation. A complete prediction-execution-verification temporal causal chain can be constructed, recording not only the physical manifestations of operations but also quantifying the deviation between subjective predictions and objective physical feedback. This improves the authenticity, verifiability, and accuracy of spatial relation representation in the physical cognitive layer corpus.

[0224] For metacognitive layer corpus units, confidence boundaries can be determined based on the expressive features of cognitive segments, and the change in confidence boundaries between adjacent cognitive segments can be calculated. Metacognitive layer corpus units can then be determined based on the change in confidence boundaries.

[0225] In some examples, confidence boundary values ​​can be calculated using the conclusion expression strength, uncertainty expression strength, correction strength, and verification strength contained in the fragment-level basic feature vectors corresponding to cognitive fragment units. The change in confidence boundary between adjacent cognitive fragment units can then be calculated to map the change in confidence boundary to the corresponding metacognitive state. To construct the quantification calculation logic, confidence boundary value formulas and boundary change formulas can be introduced.

[0226] First, to quantitatively characterize the degree of certainty in subjective thinking at the current moment, a metacognitive level rating function, i.e., the confidence boundary value formula, can be used for calculation. The confidence boundary value formula can be specifically expressed as follows:

[0227] ;

[0228] in, Indicates the time series index number is The confidence boundary value corresponding to the cognitive segment unit; This represents the strength parameter of the conclusion expression extracted from the fragment-level basic feature vector; This represents the uncertain expression intensity parameter extracted from the fragment-level basic feature vector; This represents the correction intensity parameter extracted from the fragment-level basic feature vector; This represents the proof strength parameter extracted from the fragment-level basic feature vector; This represents the pre-defined weighting constants used to balance the various expression dimensions.

[0229] Furthermore, in order to capture the dynamic evolution trend of subjective cognitive state, the boundary change formula can be used to calculate the level difference between adjacent cognitive segments. Specifically, the boundary change formula can be expressed as:

[0230] ;

[0231] in, This represents the change in the confidence margin of the current cognitive segment unit compared to its temporally adjacent preceding cognitive segment unit. Indicates the time series index number is The confidence boundary value corresponding to the previous cognitive segment unit.

[0232] The calculated change in confidence boundaries can be used to... In addition, the behavior-triggered events within the cognitive segment unit are input into preset judgment rules to determine the corresponding metacognitive state. The specific judgment rules are as follows:

[0233] Judgment Rule 1: When the change in confidence boundary satisfies At that time, among them, The first preset change threshold is used to determine the corresponding metacognitive state as a boundary contraction state, which represents the enhancement of subjective thinking confidence.

[0234] Judgment Rule 2: When the change in the confidence boundary satisfies At that time, among them, The second change threshold is a preset threshold. The corresponding metacognitive state is determined to be a boundary expansion state, which represents the expansion of subjective doubt or the diffusion of thinking space. When the change in the confidence boundary is greater than or equal to the negative second change threshold and less than or equal to the first change threshold, and no triggering event corresponding to the third or fourth judgment rule is detected, the metacognitive state corresponding to the previous cognitive segment can be maintained and the metacognitive state change is not triggered.

[0235] Judgment Rule 3: When a triggering event is detected in the original interaction data where a historical conclusion is revoked and counter-evidence is activated, the corresponding metacognitive state is determined to be a boundary rupture state, indicating that the previous cognitive framework has been subversively shaken.

[0236] Judgment Rule 4: When a triggering event is detected in which the verification operation within the current cognitive segment unit is successful and the space of alternative action is reduced, the corresponding metacognitive state is determined to be the boundary restoration state, which represents the re-establishment of subjective thinking order.

[0237] The extracted and packaged metacognitive layer corpus units can be expressed as a tuple data structure with a standardized data format:

[0238] ;

[0239] in, Indicates the time series index number is The cognitive fragment unit corresponds to the metacognitive layer corpus unit; It represents the set of core cognitive objects or target knowledge units that are locked onto by the current thought process; This represents the time span record corresponding to the corpus unit in the metacognitive layer; This represents the calculated change in the confidence boundary. Labels representing internal psychological events or external behavioral stimuli that trigger a shift in metacognitive states; It indicates a definite metacognitive state transition type, specifically including at least one of the following: boundary contraction state, boundary expansion state, boundary rupture state, and boundary recovery state; This represents a record of thought correction or cognitive reshaping strategies matched to the metacognitive state transition process.

[0240] In this embodiment, by extracting multidimensional expression features within cognitive segment units, and calculating confidence boundary values ​​based on a confidence boundary value formula that includes an addition term for the intensity of conclusion expression and a deduction term for the intensity of uncertainty, correction, and verification, and combining the confidence boundary changes of adjacent cognitive segments with interactive triggering events, when the corresponding judgment conditions of boundary contraction, boundary expansion, boundary rupture, and boundary recovery are met, metacognitive information is encapsulated into a metacognitive layer corpus unit tuple structure containing core cognitive objects, time span, boundary changes, triggering events, state transition types, and thinking correction strategies. Thus, the deep self-monitoring states such as subjective doubt, certainty, and wavering hidden behind text operations or speech intervals can be quantitatively represented, and the changes in metacognitive states can be characterized by confidence boundary changes and state judgment rules, thereby improving the accuracy of the interactive corpus in representing reflection paths and cognitive adaptation processes.

[0241] For style and capability layer corpus units, cross-task features can be extracted in a multi-task environment and style stability can be calculated. Based on style stability, style and capability layer corpus units can be determined.

[0242] Specifically, it can aggregate nine-layer extraction results corresponding to multiple different cognitive segments in time sequence as multi-task, multi-dimensional inputs for statistical benchmarks. By conducting long-term statistical analysis and tracking of cognitive indicators across different task environments, stable feature signals that exhibit preset stability characteristics in cross-task environments can be screened to form long-term style and capability fingerprints.

[0243] For example, when determining the style and capability layer corpus units, the following detailed processing flow can be executed: Extract cognitive indicator parameters obtained from multiple cognitive segment units under different business tasks. These cognitive indicator parameters may specifically include historical task execution speed, the average distribution of specific uncertainty parameters, the frequency of correction operations, and the frequency of opinion reversal when faced with conflicting evidence. Then, the extracted cognitive indicator parameters can be smoothed and subjected to variance analysis over a time span to assess the fluctuation differences of each cognitive indicator parameter across different task types. If a specific cognitive indicator parameter has a statistical variance lower than a preset stability threshold or a co-occurrence frequency greater than a preset feature emergence threshold in multiple distinct task scenarios, then that specific cognitive indicator parameter is determined as a feature that stably appears across tasks. The ratio between the number of features determined to be stably appearing across tasks and the total number of cognitive indicator parameters participating in the evaluation can be determined as style stability. When the style stability is greater than or equal to the preset style stability threshold, the style and capability layer corpus units can be generated. The style stability threshold can be determined based on the labeled historical task process data or historical sample distribution, following the method described above for determining the threshold of the scoring function. Furthermore, the selected cross-task stable features are clustered and associated to construct an individual behavioral style fingerprint and professional competence benchmark profile that can represent the specificity of subjective thinking.

[0244] The extracted and packaged style and capability layer corpus units can be expressed as a tuple structure with a standardized data format:

[0245] ;

[0246] in, This represents the style and capability corpus units determined after macro-level statistical analysis of multiple current time segments; This represents a unique identifier information that is bound to the currently observed object; This represents the generated style feature vector that reflects interaction habits, which may specifically include language preference labels, operational aggression, and pause preference parameters; This represents the set of persistent features that appear stably across tasks. This refers to the professional competence level indicators after quantitative assessment, which may include knowledge proficiency score, logical rigor rating, and complex decision response delay level. This represents the comprehensive stability trust score calculated based on the style and capability fingerprints mentioned above. The comprehensive stability trust score can be derived from the style stability.

[0247] By statistically analyzing cognitive indicators across multiple task scenarios, we selected features that consistently appear across tasks and encapsulated them into a style and competency-level corpus unit structure that includes identity identifiers, style feature vectors, a set of stable cross-task features, professional competency level indicators, and stability trust scores. This overcomes the limitations of traditional corpora that can only reflect single, occasional behaviors or local fragments, and achieves a macroscopic quantitative abstraction of the underlying behavioral styles and core competencies of subjective thinking in the long-term, multi-task evolution process. It effectively reveals the personalized deep characteristics behind surface interaction behaviors, thereby improving the accuracy and robustness of computer systems in predicting the behavior of high-order cognitive subjects in the long term, matching personalized strategies, and simulating cross-scenario competency transfer.

[0248] For the risk and normative constraint layer corpus units, sensitive risk signals, normative sources of compliance, and responsibility boundaries can be extracted to quantify and generate corpus suppression scores. The risk and normative constraint layer corpus units are then determined based on the corpus suppression scores.

[0249] Specifically, it can identify risk signals, normative sources, and responsibility boundaries within cognitive segments, and calculate corpus suppression scores to determine the manifestation of compliance constraints in the thinking and decision-making process. To construct the quantitative calculation logic, a risk and normative constraint layer scoring function, i.e., the corpus suppression score formula, can be used for dynamic quantitative calculation. The corpus suppression score formula can be specifically expressed as:

[0250] ;

[0251] in, This represents the corpus suppression score calculated for the current cognitive segment unit; Risk signal feature parameters that indicate potential safety or quality risks to identified content; This indicates the identified pre-defined normative sources or ethical and legal constraints characteristic parameters within the corresponding business scenario. These are characteristic parameters representing the degree of risk of liability boundaries involved in the identified corresponding subject's behavior, used to characterize the degree to which the responsible subject is unclear, the scope of liability conflicts, or the conditions for assuming liability are not met; This represents the pre-defined weighting constants used to balance the evaluation dimensions of various constraints.

[0252] Various risk and normative characteristic signals can be input into the above corpus suppression score formula. In the specific execution of output control and blocking determination, multi-layered dynamic critical boundaries from high to low can be set. The specific determination and control logic is as follows:

[0253] When the calculated corpus suppression score satisfies the following first constraint formula:

[0254] ;

[0255] In the above formula for the first constraint condition, This indicates a pre-set high-risk suppression threshold. When the corpus suppression score is greater than or equal to the high-risk suppression threshold, it is determined that the current interactive content has a serious safety red line risk, which can trigger the operation of suppressing output, terminating the external output of the current complete content or terminating the business response.

[0256] When the calculated corpus suppression score satisfies the following second constraint formula:

[0257] ;

[0258] In the above formula for the second constraint condition, This indicates a pre-set low-risk admission threshold; when the corpus suppression score is within the range defined by the low-risk admission threshold and the high-risk suppression threshold, it is determined that the current interactive content has a moderate potential risk and needs to be purified and desensitized, which can trigger the operation of reducing the granularity of output and selectively filter or blur sensitive information components.

[0259] The final extracted and packaged risk and normative constraint layer corpus units can be expressed as a tuple data structure with a normalized standard data format:

[0260] ;

[0261] Indicates the time series index number is The risk and normative constraint layer corpus units corresponding to the cognitive fragment units; This indicates the risk signal component belonging to the unit within this layer; This indicates the source reference index of the specification belonging to the unit at this level; This indicates the output suppression action execution status label generated based on the suppression score calculation results of the corpus; This indicates the level parameter for controlling the granularity of the interactive information output generated by dynamic matching; This refers to the set of de-identified alias texts or security alternative response schemes invoked when executing security alternative outputs; This indicates the responsibility boundary definition information belonging to the unit within this layer.

[0262] By identifying risk signals, normative sources, and responsibility boundaries in cognitive fragment units, and calculating scores based on a corpus suppression score formula that includes a weighted sum of risk signal items, normative source items, and responsibility boundary items, the system selects to perform full output, granular output, or suppressed output operations when the scores meet multi-layered critical conditions including high-risk suppression thresholds and low-risk admission thresholds. The constraint information is encapsulated into a risk and normative constraint layer corpus unit tuple structure that includes risk signals, normative sources, suppression states, granular parameters, alternative sets, and responsibility boundaries. This allows for the explicit and structured extraction and reconstruction of deep-seated implicit compliance cognitions that are known during the thinking process but are inconvenient to express directly due to objective constraints. This avoids compliance out-of-control during data generation, effectively balances cognitive openness and business security, and improves the security and credibility of interactive corpora within the computer system, the proactive prevention of behavioral boundaries, and the traceability of normative constraints.

[0263] It is not difficult to see that in the embodiments of this application, by introducing key intermediate quantitative parameters including judgment score, decision compression rate, associative jump distance, physical consequence deviation and confidence boundary change in the corpus unit determination mechanism of each cognitive layer, the traditional invisible implicit thinking trajectory is transformed into quantitative difference and state transition rules that can be directly calculated by computer, thereby improving the mathematical computability, logical consistency and objective verifiability in the cognitive corpus construction process.

[0264] Optionally, in some embodiments, determining the risk and normative constraint layer corpus unit based on the corpus suppression score includes: classifying and controlling the corpus of the corresponding cognitive segment based on the corpus suppression score.

[0265] Specifically: when the corpus suppression score is less than the low-risk admission threshold, the corpus is output in its entirety to be identified as a first-level safety unit of the risk and normative constraint layer corpus; when the corpus suppression score is greater than or equal to the low-risk admission threshold and less than the high-risk suppression threshold, the corpus is output with reduced granularity to be identified as a second-level safety unit of the risk and normative constraint layer corpus; when the corpus suppression score is greater than or equal to the high-risk suppression threshold, the corpus is output with complete suppression to be identified as a third-level safety unit of the risk and normative constraint layer corpus. In particular, when classifying and controlling the corpus of corresponding cognitive segments, three different score intervals can be pre-set to represent different safety levels. When the corpus suppression score falls within the first interval, the corpus is deemed safe and compliant. In this case, the corpus can be output without loss and is identified as a first-level safety-level risk and normative constraint layer corpus unit. When the corpus suppression score falls within the second interval, the corpus is deemed to involve moderate risk or privacy. In this case, a desensitization algorithm can be triggered to perform granular output of the corpus by blurring or abstracting it, and it is identified as a second-level safety-level risk and normative constraint layer corpus unit. When the corpus suppression score falls within the third interval, the corpus is deemed to have triggered a safety red line. In this case, a data blocking mechanism can be triggered to completely suppress the output of the corpus, prohibiting the corpus from entering the output end, and it is identified as a third-level safety-level risk and normative constraint layer corpus unit.

[0266] It is not difficult to see that in this embodiment of the application, the three-stage interval dynamic control mechanism introduces adaptive security and compliance filtering at the bottom layer of corpus construction. This can maximize the preservation of secure corpus while blocking the physical output of sensitive privacy or high-risk violation data, thereby achieving an objective balance between high corpus utilization and data security risk control.

[0267] Optionally, in some embodiments, the step of structurally representing the multi-layer cognitive corpus units to construct the multi-layer cognitive corpus, i.e., step S1062, may include:

[0268] Step S10621: Using a preset inter-layer relationship scoring function, calculate at least one of the following for any two cognitive corpus units belonging to different cognitive layers in the multi-layer cognitive corpus units: temporal proximity, citation degree, causal support degree, and content consistency degree.

[0269] Step S10622: Perform a weighted summation on at least one of the time proximity, the citation degree, the causal support degree, and the content consistency degree to obtain the inter-layer relationship score;

[0270] Step S10623: Based on the inter-layer relationship score, determine the inter-layer mapping relationship between the two cognitive corpus units to generate an inter-layer mapping relationship set;

[0271] Step S10624: The multi-layer cognitive corpus units and the set of inter-layer mapping relationships are taken as the result of the structured expression to realize the construction of the multi-layer cognitive corpus.

[0272] To achieve a network-structured representation of multi-layered cognitive corpora, cross-layered connections can be established between different cognitive corpus units, such as... Figure 4 As shown. Figure 4 In this model, K, R, J, D, A, P, M, S, and Q correspond to the knowledge layer, perception-representation layer, judgment layer, decision-making layer, association layer, physical cognition layer, metacognition layer, style and ability layer, and risk and normative constraint layer, respectively. It is important to emphasize that... Figure 4 The repeated occurrences of A or M in the diagram represent different cognitive corpus unit nodes within the same cognitive layer, and are not newly added cognitive layers. In the attached diagram, "support" corresponds to the support relationship, "trigger" corresponds to the trigger relationship, "constraint" corresponds to the constraint relationship, and "regulation" corresponds to the regulation relationship, which are used to represent the inter-layer mapping relationship between different cognitive corpus units.

[0273] Two cognitive corpus units at different cognitive levels can be arbitrarily selected, and a preset inter-level relationship scoring function can be invoked to calculate at least one of the following: temporal proximity, citation degree, causal support degree, and content consistency degree between the two cognitive corpus units at different cognitive levels.

[0274] To construct the structured quantitative calculation logic, the preset inter-layer relationship scoring function, i.e., the inter-layer relationship scoring formula, can be specifically expressed as:

[0275] ;

[0276] in, Indicates the first one drawn The first cognitive level unit With the A second cognitive level unit The inter-layer relationship score is calculated between the two, wherein the first cognitive layer unit and the second cognitive layer unit belong to different cognitive levels; This represents the temporal proximity feature signal between the first cognitive layer unit and the second cognitive layer unit in the original interaction stream; This indicates the reference degree characteristic signal generated between the two layer units due to the overlap of explicit pointers or implicit concepts; This represents the causal support feature signal obtained through calculation using a pre-causal logic tree or temporal dependency evolution. This represents the content consistency feature signal used to evaluate the semantic self-consistency of the core load information of the two layer units; This represents the pre-defined weighting constants used to balance the evaluation dimensions of inter-layer correlations.

[0277] The temporal proximity can be determined by normalizing the timestamp difference between the cognitive segments corresponding to the two cognitive corpus units; the smaller the timestamp difference, the higher the temporal proximity. The citation degree can be determined by the intersection ratio of object identifiers, concept identifiers, rule identifiers, or event identifiers contained in the two cognitive corpus units. The causal support degree can be determined by the matching results of operation triggering relationships, input-output dependencies, or preset causal rules recorded in a unified event sequence. The content consistency degree can be determined by the similarity between the text feature vectors or structured field feature vectors corresponding to the two cognitive corpus units. All of the above indicators can be normalized to the range of 0 to 1.

[0278] The weight coefficients in the inter-layer relationship scoring function can be determined by training or validation based on labeled inter-layer association samples, or they can be pre-configured according to the importance of each indicator in the current application scenario and stored in the inter-layer mapping parameter configuration table.

[0279] After obtaining the calculated dimensional indicators, a weighted summation operation can be performed on at least one of the calculated temporal proximity, citation degree, causal support degree, and content consistency degree to obtain the inter-layer relationship score. Based on the inter-layer relationship score, it can be determined whether there is a correlation between two cognitive corpus units, and then the inter-layer mapping relationship between the two cognitive corpus units can be determined to generate an inter-layer mapping relationship set. The inter-layer mapping relationship set is used as the result of structured expression to realize the construction of multi-layer cognitive corpus.

[0280] In some cases, combined Figure 5 As shown, the multi-layer cognitive corpus unit set It can be represented as:

[0281] ;

[0282] in, Represents a set of knowledge-layer corpus units. Represents a set of units in the perception-representation layer corpus. Represents the set of corpus units in the judgment layer. Represents the set of decision-making level corpus units, Represents the set of associative layer corpus units, Represents a set of corpus units at the physical cognitive level. Represents a set of corpus units at the metacognitive level. The set of corpus units representing style and ability levels. This represents the set of corpus units representing the risk and normative constraint layers.

[0283] The set of inter-layer mapping relationships is denoted as It is used to store the mapping edges between different cognitive layer corpus units and their mapping edge types; the global metadata descriptor is denoted as It is used to record the interaction background, industry domain, timestamp information, and basic static features. , and This serves as the foundational data for the subsequent unified encapsulation and formation of the Unified Cognitive Corpus (UCC).

[0284] Figure 5 The “hierarchical set” in the text corresponds to the described "Mapping set (cross-layer mapping)" corresponds to the above "Metadata" corresponds to the "Index" corresponds to the "Completeness score" corresponds to the above "Coupling score" corresponds to the above “UCC” corresponds to the Unified Cognitive Corpus structure in this application.

[0285] Optionally, in some embodiments, based on the inter-layer relationship score, an inter-layer mapping relationship is determined between the two cognitive corpus units to generate an inter-layer mapping relationship set, i.e., step S10623 may include:

[0286] Step A1: When the inter-layer relationship score is greater than or equal to a preset mapping threshold, a mapping edge is established between the two cognitive corpus units.

[0287] Step B1: Label the mapping edge type of the mapping edge according to at least one of the time proximity, the reference degree, the causal support degree, and the content consistency degree;

[0288] Step C1: Generate the set of inter-layer mapping relationships based on the mapping edges and the mapping edge types.

[0289] Specifically, a preset mapping threshold can be used to filter the inter-layer relationship score. The calculated inter-layer relationship score satisfies the following set mapping edge establishment condition formula:

[0290] ;

[0291] In the above formula, This represents a pre-set inter-layer mapping threshold. When the inter-layer relationship score is greater than or equal to the inter-layer mapping threshold, it is determined that there is a logical collaborative or evolutionary relationship between the first cognitive layer unit and the second cognitive layer unit that satisfies the preset association conditions, thereby triggering the operation of establishing a mapping edge.

[0292] After establishing the mapping edges, we can trace the dimension features that contribute the most to the inter-layer relationship score, that is, label the mapping edge type according to at least one of time proximity, citation degree, causal support degree and content consistency degree.

[0293] For example, when citation degree is the primary contribution dimension, and the second cognitive layer corpus unit cites object identifiers, concept identifiers, rule identifiers, or event identifiers in the first cognitive layer corpus unit, the mapping edge can be labeled as a support relationship; when causal support degree is the primary contribution dimension, and the event corresponding to the first cognitive layer corpus unit precedes and triggers the event corresponding to the second cognitive layer corpus unit in time sequence, the mapping edge can be labeled as a triggering relationship; when one cognitive layer corpus unit limits the output range, allowed conditions, or prohibited conditions of another cognitive layer corpus unit, the mapping edge can be labeled as a constraint relationship; when a state change corresponding to one cognitive layer corpus unit causes an adjustment to the scoring parameters, judgment thresholds, or output state of another cognitive layer corpus unit, the mapping edge can be labeled as a regulatory relationship. If two cognitive corpus units simultaneously satisfy the judgment conditions corresponding to multiple relationship types, multiple mapping edge types can be labeled for the same mapping edge.

[0294] The mapping edge can be specifically expressed as a tuple data structure with a normalized standard data format as follows:

[0295] ;

[0296] in, Indicates connection to the first cognitive layer unit With the second cognitive layer unit Specific cross-layer mapping edges; The topology type of the cross-layer mapping edge in the annotation can be at least one of the following: support relationship, trigger relationship, constraint relationship, and regulation relationship. This represents the association strength weight score assigned to the cross-layer mapping edge, and its value can be compared with the calculated inter-layer relationship score. Maintain a positive correlation.

[0297] The relationship data is compiled based on the established mapping edges and the labeled mapping edge types to generate a set of inter-layer mapping relationships.

[0298] It is not difficult to see that in the embodiments of this application, by constructing an inter-layer relationship scoring function based on temporal proximity, citation degree, causal support degree and content consistency degree, and establishing cross-layer mapping edges between the originally isolated layer units, the fragmented cognitive data is woven into a network, clearly restoring the lateral weighing trajectory and vertical evolution path of subjective thinking when facing conflicting information, which significantly improves the causal interpretability and full-flow traceability of the entire human-computer interaction cognitive process.

[0299] Optionally, in some embodiments, the step of using the multi-layer cognitive corpus units and the set of inter-layer mapping relationships as the result of the structured expression to realize the construction of the multi-layer cognitive corpus, i.e., step S10624 may include:

[0300] Step A2: The multi-layer cognitive corpus units and the set of inter-layer mapping relationships are uniformly encapsulated into a unified cognitive corpus structure;

[0301] Step B2: Calculate the corpus completeness and cross-layer coupling degree of the unified cognitive corpus structure;

[0302] Step C2: Based on the completeness of the corpus and the cross-layer coupling degree, control the output quality of the unified cognitive corpus structure to output multi-layer cognitive corpus results that meet the quality standards.

[0303] Specifically, the nine layers of cognitive corpus units (as data node metadata) extracted independently in the aforementioned steps and the generated set of inter-layer mapping relationships (as topological connection relationship data) can be merged and uniformly packaged into a unified cognitive corpus structure, thereby forming a multi-dimensional integrated corpus data cube.

[0304] To achieve standardized and structured integrated encapsulation, the unified cognitive corpus structure can be specifically expressed as the following tuple encapsulation structure:

[0305] ;

[0306] in, This represents the structured data object of the final integrated unified cognitive corpus. This represents a global metadata descriptor used to record interaction background, industry domain, timestamp information, and basic static characteristics; It represents a collection of multi-layered cognitive corpus units composed of extracted knowledge layer, perception-representation layer, judgment layer, decision-making layer, association layer, physical cognition layer, metacognition layer, style and ability layer, and risk and normative constraint layer; This represents the set of cross-layer mapping relationships generated through the aforementioned inter-layer relationship scoring and mapping edge establishment process; This refers to an inverted index or multidimensional spatial directory identifier that is pre-built for each layer of data nodes and mapping edges in order to improve retrieval efficiency; The corpus completeness control score is used to quantitatively evaluate the overall information richness of the structure of a specific unified cognitive corpus. This represents the cross-layer coupling control score, which is used to quantify the logical convergence and degree of correlation between cascaded networks of different cognitive levels.

[0307] After encapsulation, the output quality of the unified cognitive corpus structure can be verified. Specifically, the corpus completeness of the unified cognitive corpus structure can be calculated to check for missing data at each layer. To achieve objective structured quality verification, in some examples, the corpus completeness can be calculated using the following indicator function summation formula:

[0308] ;

[0309] In the above formula, This represents the normalized corpus completeness control score calculated for the unified cognitive corpus structure. Indicates the first A set of cognitive corpus units corresponding to each cognitive level; The not-equals logical operator; Represents the empty set; This indicates an indicator function that outputs 1 when the corresponding cognitive level is not empty and 0 when the corresponding cognitive level is empty. This indicates the total number of preset cognitive layers in the current multi-layer cognitive corpus; in an embodiment using a nine-layer cognitive corpus unit, .

[0310] Simultaneously, the cross-layer coupling degree of the unified cognitive corpus structure can be further calculated to check for anomalies such as excessively sparse or excessively dense connections between layers. The cross-layer coupling degree can be calculated using the following formula for average cross-layer connectivity:

[0311] ;

[0312] In the above formula, This represents the cross-layer coupling control score calculated for the structure of the unified cognitive corpus; Represents the set of inter-layer mapping relationships The absolute number of valid cross-layer mapping edges successfully established in the middle; Indicates the first The absolute number of independent corpus unit nodes contained in a specific cognitive level; Indicates all The total number of cognitive corpus unit nodes contained in each cognitive level; Indicates the total number of cognitive layers; This represents the summation index at the cognitive level; This represents a small positive smoothing fault-tolerance constant used to prevent the denominator from being zero; for example, it can be taken as... .

[0313] Based on the calculated corpus completeness and cross-layer coupling degree, the output quality of the unified cognitive corpus structure is controlled. When both the corpus completeness and cross-layer coupling degree reach the preset quality standards (e.g., the corpus completeness is greater than or equal to the set completeness threshold, and the cross-layer coupling degree is within the preset reasonable density range), the output is allowed to proceed, so as to output multi-layer cognitive corpus results that meet the quality standards.

[0314] It is not difficult to see that in the embodiments of this application, the nine layers of independently extracted corpus node data and inter-layer mapping relationship data are multidimensionally integrated, encapsulated and output quality verified by a unified cognitive corpus structure, forming a data structure with standard index and quality verification results. This improves the reusability, retrieval efficiency, cross-scenario migration capability and output quality of advanced cognitive corpus in heterogeneous computer systems or different business migration scenarios.

[0315] Figure 6 This illustration shows yet another exemplary flowchart of the multi-layered cognitive corpus construction method according to an embodiment of this application. Figure 6 As shown, the input is the original multi-source task process data, which includes at least two of the following: audio perception data, text interaction data, behavioral operation data, environmental state data, and verification feedback data. Steps 1-2 are executed to construct a unified event sequence (corresponding to the unified event flow construction in the figure). Step 3 is executed to segment cognitive segments to generate multiple context-stable cognitive segments. Step 4 is executed to extract segment-level basic feature vectors. Steps 5-13 are executed to extract nine-layer cognitive corpus units, that is, using independent scoring functions and judgment conditions established for each cognitive layer, multi-layer cognitive corpus units are extracted from each cognitive segment and its corresponding common feature matrix. Step 14 is executed to construct cross-layer mapping, that is, using a preset inter-layer relationship scoring function to generate a set of inter-layer mapping relationships. Step 15 is executed to perform unified encapsulation, that is, to encapsulate the multi-layer cognitive corpus units and the set of inter-layer mapping relationships into a unified cognitive corpus structure. The final output is the Unified Cognitive Corpus (UCC) shown in the attached figure (corresponding to the unified cognitive corpus structure in this application).

[0316] The steps of the various methods described above are only for clarity. In practice, they can be combined into one step or some steps can be split into multiple steps. As long as they include the same logical relationship, they are all within the scope of protection of this application. Adding insignificant modifications or introducing insignificant designs to the algorithm or process, but without changing the core design of the algorithm and process, are also within the scope of protection of this application.

[0317] Furthermore, some embodiments of this application also provide an electronic device. The electronic device can be various forms of digital computer, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, mainframe computers, cellular phones, smartphones, wearable devices, and other similar computing devices.

[0318] The electronic device includes: one or more processors; and a memory storing computer program instructions, which, when executed, cause the processor to perform the steps of the methods provided in any one or more of the above embodiments.

[0319] Figure 7 An exemplary structural diagram of the electronic device is disclosed. The electronic device includes one or more processors 1101, a memory 1102, an input device 1103, and an output device 1104. The various components are interconnected via a bus or other means (the diagram shows an example of bus connection). The processor 1101 can be used to execute instructions stored in the memory 1102 to control the overall operation of the electronic device. The memory 1102 may include a program storage area and a data storage area, wherein the program storage area stores the operating system and applications required for at least one function; the data storage area stores data created according to the use of the electronic device, etc. The memory 1102 may include high-speed random access memory and may also include non-transitory memory, such as disk storage devices, flash memory devices, or other non-transitory solid-state storage devices. In some embodiments, the memory 1102 may also include storage resources located remotely to the processor and accessible via a network.

[0320] Input device 1103 can be used to receive input numerical or character information or user operation signals, such as a touch screen, keypad, mouse, trackpad, touchpad, indicator, one or more mouse buttons, trackball, joystick, etc. Output device 1104 may include display devices (such as liquid crystal displays, light-emitting diode displays, plasma displays, and optional touch screens), auxiliary lighting devices (such as LEDs), and haptic feedback devices (such as vibration motors), etc.

[0321] To facilitate user interaction, the electronic device may be configured to include a display device (such as an LCD or CRT monitor) and input devices such as a keyboard and pointing devices (e.g., a mouse or touchpad). Feedback can be any form of sensory feedback (e.g., visual feedback, auditory feedback); input may also be received via voice, touch, or other means.

[0322] This application also relates to a computer-readable medium storing computer program instructions thereon, which, when executed by a processor, implement the steps of the methods provided in any one or more of the above embodiments. The computer-readable medium may be a memory included in an electronic device, or it may be a standalone storage medium not assembled into the device.

[0323] It should be noted that the computer-readable medium described in this application may be a computer-readable signal medium, a computer-readable storage medium, or a combination of both. Examples include, but are not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. Specific examples of storage media may include, but are not limited to, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory, optical fibers, portable CD-ROMs, optical storage devices, magnetic storage devices, etc., or any suitable combination thereof.

[0324] Computer-readable media may store one or more programs that can be used by or in conjunction with an instruction execution system. The media may be permanent or non-permanent, removable or non-removable, and may store information by any method or technology, including computer-readable instructions, data structures, program modules, or other data.

[0325] The computer program code used to implement the operations of this application can be written in any combination of one or more programming languages, including object-oriented programming languages ​​(such as Java, Smalltalk, and C++) and conventional procedural programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer, partially on a remote computer, or entirely on a remote computer or server. The remote computer can be connected to the user's computer via any network (including a local area network or a wide area network) or can be connected to an external computer.

[0326] In the above embodiments, the functions can be implemented in whole or in part by software, hardware, firmware, or any combination thereof, for example, by using application-specific integrated circuits, general-purpose computers, or other similar hardware devices. In some embodiments, the software program of this application can be executed by a processor to implement the steps or functions; it can also be implemented by hardware, for example, as a circuit that works in conjunction with the processor to execute the steps or functions.

[0327] This application also provides a computer program product, including one or more computer program instructions. When executed by a processor, these computer program instructions, in whole or in part, generate the processes or functions described in this application. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium or transmitted from one storage medium to another via wired (e.g., DSL) or wireless (e.g., wireless, microwave) means. The computer-readable storage medium may be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive).

[0328] The flowcharts or block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of devices, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-specific system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0329] The scope of this application is defined by the appended claims rather than the foregoing description, and is therefore intended to encompass all variations falling within the meaning and scope of equivalents of the claims. No reference numerals in the claims should be construed as limiting the scope of the claims. Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a device claim may also be implemented by a single unit or device in software or hardware. Terms such as "first," "second," etc., are used only for distinguishing descriptions and do not indicate any particular order, nor should they be construed as indicating or implying relative importance.

[0330] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any corresponding changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims, and the above embodiments should be regarded as exemplary and non-limiting.

Claims

1. A method for constructing a multi-layered cognitive corpus, characterized in that, The method includes: Based on the original multi-source task process data, determine the state characteristics of each modality candidate segment; Based on the aforementioned state characteristics, candidate segments that meet the triggering conditions are identified as standard events to form a unified event sequence; Based on the dynamic mutation characteristics of the unified event sequence in a preset dimension, the segmentation boundary of the unified event sequence is determined to generate multiple context-stable cognitive segments. Multidimensional features are extracted for each cognitive segment to obtain a segment-level basic feature vector; Construct a common feature matrix based on the fragment-level basic feature vectors; Based on the shared feature matrix, the construction of the multi-layer cognitive corpus is realized; The step of determining candidate segments that meet the triggering conditions as standard events based on the state features to form a unified event sequence includes: using event triggering functions corresponding to preset event triggering rules to determine the state features of each modal candidate segment; wherein, the preset event triggering rules include at least one of speech pause rules, text editing behavior determination rules, atomic operation recognition rules, object state change rules, and verification feedback triggering rules; when a certain state feature meets the corresponding event triggering function, the candidate segment to which the state feature belongs is converted into a standard event; and all the generated standard events are sorted according to... The timestamps are arranged in chronological order, and the arranged standard event stream is divided into local alignment windows. Within the local alignment window, if any two standard events meet the set time difference condition, the content relationship and object overlap relationship between the two standard events are determined. When the two standard events simultaneously meet the content similarity condition and the object overlap condition, the two standard events are repeatedly merged. When the two standard events meet the object overlap condition, and their content relationship indicates opposite operation types, state change directions, or semantic conclusions, the two standard events are conflict-splitting processed to generate the unified event sequence with multimodal alignment. The preset dimensions include at least one of the following: time interval dimension, semantic content dimension, task stage dimension, and feedback or risk status dimension. The step of determining the segmentation boundaries of the unified event sequence based on the dynamic mutation characteristics of the unified event sequence in the preset dimensions to generate multiple context-stable cognitive segments includes: extracting feature signals corresponding to each dimension included in the preset dimensions; calculating a comprehensive boundary judgment value based on the feature signals; when the comprehensive boundary judgment value indicates that at least one dimension in the preset dimensions has undergone a dynamic mutation, determining the corresponding event position as the segmentation boundary of the unified event sequence; and truncating the sequence data between two adjacent segmentation boundaries in the unified event sequence based on the segmentation boundaries to generate multiple context-stable cognitive segments. The construction of the multi-layer cognitive corpus based on the shared feature matrix includes: establishing scoring functions and judgment conditions for each of the multiple cognitive layers constituting the multi-layer cognitive corpus; extracting multi-layer cognitive corpus units based on each cognitive segment and the feature data corresponding to that cognitive segment in the shared feature matrix; and performing structured expression on the multi-layer cognitive corpus units to realize the construction of the multi-layer cognitive corpus.

2. The method according to claim 1, characterized in that, The original multi-source task process data includes at least two of the following: audio perception data, text interaction data, behavioral operation data, environmental state data, and verification feedback data. The process of determining the state features of each modality candidate segment based on the original multi-source task process data includes: If the original multi-source task process data includes the audio perception data, then based on the audio perception data, extract the state features of the sentence-level segments as the candidate segments; If the original multi-source task process data includes the text interaction data, then based on the text interaction data, extract the state features of the edited segment as the candidate segment; If the original multi-source task process data includes the behavioral operation data, then based on the behavioral operation data, extract the state features of the atomic operation fragments as the candidate fragments; If the original multi-source task process data includes the environmental state data, then based on the environmental state data, extract the state features of the state change segments as the candidate segments; If the original multi-source task process data includes the verification feedback data, then the state features of the feedback fragments as candidate fragments are extracted based on the verification feedback data.

3. The method according to claim 1, characterized in that, The multi-layer cognitive corpus unit specifically comprises a nine-layer cognitive corpus unit; The nine-layer cognitive corpus unit includes: knowledge layer corpus unit, perception-representation layer corpus unit, judgment layer corpus unit, decision layer corpus unit, association layer corpus unit, physical cognition layer corpus unit, metacognition layer corpus unit, style and ability layer corpus unit, and risk and normative constraint layer corpus unit.

4. The method according to claim 3, characterized in that, The step involves establishing scoring functions and judgment conditions for each of the multiple cognitive layers constituting the multi-layered cognitive corpus. Based on each cognitive segment and the feature data corresponding to that cognitive segment in the shared feature matrix, multi-layered cognitive corpus units are extracted, including: The knowledge layer corpus units are determined based on the knowledge retrieval intensity of each knowledge unit and the set knowledge layer threshold; The perception-representation layer corpus unit is determined based on the focus switching score and the set perception threshold; the focus switching score is determined based on at least one of object change features, gaze or trajectory offset features, semantic focus deviation features, and interface control switching features. The judgment layer corpus unit is determined based on the judgment score, the set validity threshold, the set flip boundary, and the exclusion conditions; the judgment score is determined based on positive evidence and counter-evidence. The decision-making layer corpus units are determined based on the decision compression rate; the decision compression rate is calculated based on candidate actions and excluded actions. The association layer corpus unit is determined based on the association trigger score and the set association threshold; the association trigger score is determined based on concept jump events. The physical consequence deviation is used to determine the physical cognition layer corpus unit; the physical consequence deviation is calculated based on the alignment of predicted consequences and actual consequences in the physical cognition elements. The metacognitive layer corpus unit is determined based on the change in confidence boundaries; the change in confidence boundaries is determined based on expressive features and adjacent cognitive segments. The style and capability layer corpus units are determined based on style stability; the style stability is determined based on cross-task features reflecting multi-task scenarios. The risk and normative constraint layer corpus units are determined based on the corpus suppression score; the corpus suppression score is determined based on risk signals, normative sources, and responsibility boundaries.

5. The method according to claim 4, characterized in that, The step of determining the risk and normative constraint layer corpus units based on the corpus suppression score includes: Based on the corpus suppression score, the corpus corresponding to the cognitive segment is classified and controlled; wherein, a low-risk admission threshold and a high-risk suppression threshold are preset, and the low-risk admission threshold is less than the high-risk suppression threshold. When the corpus suppression score is less than the low-risk admission threshold, the corpus is output in its entirety to be identified as the risk and normative constraint layer corpus unit of the first security level. When the corpus suppression score is greater than or equal to the low-risk admission threshold and less than the high-risk suppression threshold, the corpus is degranulated and output to determine the risk and normative constraint layer corpus unit as the second security level. When the corpus suppression score is greater than or equal to the high-risk suppression threshold, the corpus is completely suppressed and output to determine the risk and normative constraint layer corpus unit as the third safety level.

6. The method according to claim 1, characterized in that, The process of structuring the units of the multi-layered cognitive corpus to construct the multi-layered cognitive corpus includes: Using a preset inter-layer relationship scoring function, for any two cognitive corpus units belonging to different cognitive layers in the multi-layer cognitive corpus units, at least one of temporal proximity, citation degree, causal support degree and content consistency degree is calculated; The inter-layer relationship score is obtained by weighted summation of at least one of the time proximity, the citation degree, the causal support degree, and the content consistency degree. Based on the inter-layer relationship score, an inter-layer mapping relationship is determined between the two cognitive corpus units to generate an inter-layer mapping relationship set; The multi-layer cognitive corpus units and the set of inter-layer mapping relationships are used as the results of the structured expression to realize the construction of the multi-layer cognitive corpus.

7. The method according to claim 6, characterized in that, Based on the inter-layer relationship score, an inter-layer mapping relationship is determined between the two cognitive corpus units to generate a set of inter-layer mapping relationships, including: When the inter-layer relationship score is greater than or equal to a preset mapping threshold, a mapping edge is established between the two cognitive corpus units; The mapping edge type is labeled according to at least one of the time proximity, the reference degree, the causal support degree, and the content consistency degree; The set of inter-layer mapping relationships is generated based on the mapping edges and the mapping edge types.

8. The method according to claim 6, characterized in that, The construction of the multi-layer cognitive corpus, which uses the multi-layer cognitive corpus units and the set of inter-layer mapping relationships as the result of the structured expression, includes: The multi-layer cognitive corpus units and the set of inter-layer mapping relationships are uniformly encapsulated into a unified cognitive corpus structure; Calculate the corpus completeness and cross-layer coupling degree of the unified cognitive corpus structure; Based on the completeness of the corpus and the degree of cross-layer coupling, the output quality of the unified cognitive corpus structure is controlled to output multi-layer cognitive corpus results that meet quality standards.

9. An electronic device, characterized in that, The electronic device includes: One or more processors; and A memory storing computer program instructions, which, when executed, cause the processor to perform the steps of the method as described in any one of claims 1 to 8.

10. A computer-readable medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 8.

11. A computer program product comprising computer program instructions, characterized in that, When executed by a processor, the computer program instructions implement the steps of the method according to any one of claims 1 to 8.

Images